fs/btrfs/disk-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/radix-tree.h>
   9 #include <linux/writeback.h>
  10 #include <linux/buffer_head.h>
  11 #include <linux/workqueue.h>
  12 #include <linux/kthread.h>
  13 #include <linux/slab.h>
  14 #include <linux/migrate.h>
  15 #include <linux/ratelimit.h>
  16 #include <linux/uuid.h>
  17 #include <linux/semaphore.h>
  18 #include <linux/error-injection.h>
  19 #include <linux/crc32c.h>
  20 #include <linux/sched/mm.h>
  21 #include <asm/unaligned.h>
  22 #include "ctree.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "btrfs_inode.h"
  26 #include "volumes.h"
  27 #include "print-tree.h"
  28 #include "locking.h"
  29 #include "tree-log.h"
  30 #include "free-space-cache.h"
  31 #include "free-space-tree.h"
  32 #include "inode-map.h"
  33 #include "check-integrity.h"
  34 #include "rcu-string.h"
  35 #include "dev-replace.h"
  36 #include "raid56.h"
  37 #include "sysfs.h"
  38 #include "qgroup.h"
  39 #include "compression.h"
  40 #include "tree-checker.h"
  41 #include "ref-verify.h"
  42
  43 #ifdef CONFIG_X86
  44 #include <asm/cpufeature.h>
  45 #endif
  46
  47 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
  48                                  BTRFS_HEADER_FLAG_RELOC |\
  49                                  BTRFS_SUPER_FLAG_ERROR |\
  50                                  BTRFS_SUPER_FLAG_SEEDING |\
  51                                  BTRFS_SUPER_FLAG_METADUMP |\
  52                                  BTRFS_SUPER_FLAG_METADUMP_V2)
  53
  54 static const struct extent_io_ops btree_extent_io_ops;
  55 static void end_workqueue_fn(struct btrfs_work *work);
  56 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  57 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  58                                       struct btrfs_fs_info *fs_info);
  59 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  60 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
  61                                         struct extent_io_tree *dirty_pages,
  62                                         int mark);
  63 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
  64                                        struct extent_io_tree *pinned_extents);
  65 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
  66 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
  67
  68 /*
  69  * btrfs_end_io_wq structs are used to do processing in task context when an IO
  70  * is complete.  This is used during reads to verify checksums, and it is used
  71  * by writes to insert metadata for new file extents after IO is complete.
  72  */
  73 struct btrfs_end_io_wq {
  74         struct bio *bio;
  75         bio_end_io_t *end_io;
  76         void *private;
  77         struct btrfs_fs_info *info;
  78         blk_status_t status;
  79         enum btrfs_wq_endio_type metadata;
  80         struct btrfs_work work;
  81 };
  82
  83 static struct kmem_cache *btrfs_end_io_wq_cache;
  84
  85 int __init btrfs_end_io_wq_init(void)
  86 {
  87         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
  88                                         sizeof(struct btrfs_end_io_wq),
  89                                         0,
  90                                         SLAB_MEM_SPREAD,
  91                                         NULL);
  92         if (!btrfs_end_io_wq_cache)
  93                 return -ENOMEM;
  94         return 0;
  95 }
  96
  97 void __cold btrfs_end_io_wq_exit(void)
  98 {
  99         kmem_cache_destroy(btrfs_end_io_wq_cache);
 100 }
 101
 102 /*
 103  * async submit bios are used to offload expensive checksumming
 104  * onto the worker threads.  They checksum file and metadata bios
 105  * just before they are sent down the IO stack.
 106  */
 107 struct async_submit_bio {
 108         void *private_data;
 109         struct bio *bio;
 110         extent_submit_bio_start_t *submit_bio_start;
 111         int mirror_num;
 112         /*
 113          * bio_offset is optional, can be used if the pages in the bio
 114          * can't tell us where in the file the bio should go
 115          */
 116         u64 bio_offset;
 117         struct btrfs_work work;
 118         blk_status_t status;
 119 };
 120
 121 /*
 122  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 123  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 124  * the level the eb occupies in the tree.
 125  *
 126  * Different roots are used for different purposes and may nest inside each
 127  * other and they require separate keysets.  As lockdep keys should be
 128  * static, assign keysets according to the purpose of the root as indicated
 129  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
 130  * roots have separate keysets.
 131  *
 132  * Lock-nesting across peer nodes is always done with the immediate parent
 133  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 134  * subclass to avoid triggering lockdep warning in such cases.
 135  *
 136  * The key is set by the readpage_end_io_hook after the buffer has passed
 137  * csum validation but before the pages are unlocked.  It is also set by
 138  * btrfs_init_new_buffer on freshly allocated blocks.
 139  *
 140  * We also add a check to make sure the highest level of the tree is the
 141  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 142  * needs update as well.
 143  */
 144 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 145 # if BTRFS_MAX_LEVEL != 8
 146 #  error
 147 # endif
 148
 149 static struct btrfs_lockdep_keyset {
 150         u64                     id;             /* root objectid */
 151         const char              *name_stem;     /* lock name stem */
 152         char                    names[BTRFS_MAX_LEVEL + 1][20];
 153         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 154 } btrfs_lockdep_keysets[] = {
 155         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 156         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 157         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 158         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 159         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 160         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 161         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 162         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 163         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 164         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 165         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 166         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
 167         { .id = 0,                              .name_stem = "tree"     },
 168 };
 169
 170 void __init btrfs_init_lockdep(void)
 171 {
 172         int i, j;
 173
 174         /* initialize lockdep class names */
 175         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 176                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 177
 178                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 179                         snprintf(ks->names[j], sizeof(ks->names[j]),
 180                                  "btrfs-%s-%02d", ks->name_stem, j);
 181         }
 182 }
 183
 184 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 185                                     int level)
 186 {
 187         struct btrfs_lockdep_keyset *ks;
 188
 189         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 190
 191         /* find the matching keyset, id 0 is the default entry */
 192         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 193                 if (ks->id == objectid)
 194                         break;
 195
 196         lockdep_set_class_and_name(&eb->lock,
 197                                    &ks->keys[level], ks->names[level]);
 198 }
 199
 200 #endif
 201
 202 /*
 203  * extents on the btree inode are pretty simple, there's one extent
 204  * that covers the entire device
 205  */
 206 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
 207                 struct page *page, size_t pg_offset, u64 start, u64 len,
 208                 int create)
 209 {
 210         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 211         struct extent_map_tree *em_tree = &inode->extent_tree;
 212         struct extent_map *em;
 213         int ret;
 214
 215         read_lock(&em_tree->lock);
 216         em = lookup_extent_mapping(em_tree, start, len);
 217         if (em) {
 218                 em->bdev = fs_info->fs_devices->latest_bdev;
 219                 read_unlock(&em_tree->lock);
 220                 goto out;
 221         }
 222         read_unlock(&em_tree->lock);
 223
 224         em = alloc_extent_map();
 225         if (!em) {
 226                 em = ERR_PTR(-ENOMEM);
 227                 goto out;
 228         }
 229         em->start = 0;
 230         em->len = (u64)-1;
 231         em->block_len = (u64)-1;
 232         em->block_start = 0;
 233         em->bdev = fs_info->fs_devices->latest_bdev;
 234
 235         write_lock(&em_tree->lock);
 236         ret = add_extent_mapping(em_tree, em, 0);
 237         if (ret == -EEXIST) {
 238                 free_extent_map(em);
 239                 em = lookup_extent_mapping(em_tree, start, len);
 240                 if (!em)
 241                         em = ERR_PTR(-EIO);
 242         } else if (ret) {
 243                 free_extent_map(em);
 244                 em = ERR_PTR(ret);
 245         }
 246         write_unlock(&em_tree->lock);
 247
 248 out:
 249         return em;
 250 }
 251
 252 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
 253 {
 254         return crc32c(seed, data, len);
 255 }
 256
 257 void btrfs_csum_final(u32 crc, u8 *result)
 258 {
 259         put_unaligned_le32(~crc, result);
 260 }
 261
 262 /*
 263  * compute the csum for a btree block, and either verify it or write it
 264  * into the csum field of the block.
 265  */
 266 static int csum_tree_block(struct btrfs_fs_info *fs_info,
 267                            struct extent_buffer *buf,
 268                            int verify)
 269 {
 270         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 271         char result[BTRFS_CSUM_SIZE];
 272         unsigned long len;
 273         unsigned long cur_len;
 274         unsigned long offset = BTRFS_CSUM_SIZE;
 275         char *kaddr;
 276         unsigned long map_start;
 277         unsigned long map_len;
 278         int err;
 279         u32 crc = ~(u32)0;
 280
 281         len = buf->len - offset;
 282         while (len > 0) {
 283                 /*
 284                  * Note: we don't need to check for the err == 1 case here, as
 285                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
 286                  * and 'min_len = 32' and the currently implemented mapping
 287                  * algorithm we cannot cross a page boundary.
 288                  */
 289                 err = map_private_extent_buffer(buf, offset, 32,
 290                                         &kaddr, &map_start, &map_len);
 291                 if (err)
 292                         return err;
 293                 cur_len = min(len, map_len - (offset - map_start));
 294                 crc = btrfs_csum_data(kaddr + offset - map_start,
 295                                       crc, cur_len);
 296                 len -= cur_len;
 297                 offset += cur_len;
 298         }
 299         memset(result, 0, BTRFS_CSUM_SIZE);
 300
 301         btrfs_csum_final(crc, result);
 302
 303         if (verify) {
 304                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 305                         u32 val;
 306                         u32 found = 0;
 307                         memcpy(&found, result, csum_size);
 308
 309                         read_extent_buffer(buf, &val, 0, csum_size);
 310                         btrfs_warn_rl(fs_info,
 311                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
 312                                 fs_info->sb->s_id, buf->start,
 313                                 val, found, btrfs_header_level(buf));
 314                         return -EUCLEAN;
 315                 }
 316         } else {
 317                 write_extent_buffer(buf, result, 0, csum_size);
 318         }
 319
 320         return 0;
 321 }
 322
 323 /*
 324  * we can't consider a given block up to date unless the transid of the
 325  * block matches the transid in the parent node's pointer.  This is how we
 326  * detect blocks that either didn't get written at all or got written
 327  * in the wrong place.
 328  */
 329 static int verify_parent_transid(struct extent_io_tree *io_tree,
 330                                  struct extent_buffer *eb, u64 parent_transid,
 331                                  int atomic)
 332 {
 333         struct extent_state *cached_state = NULL;
 334         int ret;
 335         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
 336
 337         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 338                 return 0;
 339
 340         if (atomic)
 341                 return -EAGAIN;
 342
 343         if (need_lock) {
 344                 btrfs_tree_read_lock(eb);
 345                 btrfs_set_lock_blocking_read(eb);
 346         }
 347
 348         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 349                          &cached_state);
 350         if (extent_buffer_uptodate(eb) &&
 351             btrfs_header_generation(eb) == parent_transid) {
 352                 ret = 0;
 353                 goto out;
 354         }
 355         btrfs_err_rl(eb->fs_info,
 356                 "parent transid verify failed on %llu wanted %llu found %llu",
 357                         eb->start,
 358                         parent_transid, btrfs_header_generation(eb));
 359         ret = 1;
 360
 361         /*
 362          * Things reading via commit roots that don't have normal protection,
 363          * like send, can have a really old block in cache that may point at a
 364          * block that has been freed and re-allocated.  So don't clear uptodate
 365          * if we find an eb that is under IO (dirty/writeback) because we could
 366          * end up reading in the stale data and then writing it back out and
 367          * making everybody very sad.
 368          */
 369         if (!extent_buffer_under_io(eb))
 370                 clear_extent_buffer_uptodate(eb);
 371 out:
 372         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 373                              &cached_state);
 374         if (need_lock)
 375                 btrfs_tree_read_unlock_blocking(eb);
 376         return ret;
 377 }
 378
 379 /*
 380  * Return 0 if the superblock checksum type matches the checksum value of that
 381  * algorithm. Pass the raw disk superblock data.
 382  */
 383 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
 384                                   char *raw_disk_sb)
 385 {
 386         struct btrfs_super_block *disk_sb =
 387                 (struct btrfs_super_block *)raw_disk_sb;
 388         u16 csum_type = btrfs_super_csum_type(disk_sb);
 389         int ret = 0;
 390
 391         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 392                 u32 crc = ~(u32)0;
 393                 char result[sizeof(crc)];
 394
 395                 /*
 396                  * The super_block structure does not span the whole
 397                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 398                  * is filled with zeros and is included in the checksum.
 399                  */
 400                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 401                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 402                 btrfs_csum_final(crc, result);
 403
 404                 if (memcmp(raw_disk_sb, result, sizeof(result)))
 405                         ret = 1;
 406         }
 407
 408         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 409                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
 410                                 csum_type);
 411                 ret = 1;
 412         }
 413
 414         return ret;
 415 }
 416
 417 static int verify_level_key(struct btrfs_fs_info *fs_info,
 418                             struct extent_buffer *eb, int level,
 419                             struct btrfs_key *first_key, u64 parent_transid)
 420 {
 421         int found_level;
 422         struct btrfs_key found_key;
 423         int ret;
 424
 425         found_level = btrfs_header_level(eb);
 426         if (found_level != level) {
 427 #ifdef CONFIG_BTRFS_DEBUG
 428                 WARN_ON(1);
 429                 btrfs_err(fs_info,
 430 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
 431                           eb->start, level, found_level);
 432 #endif
 433                 return -EIO;
 434         }
 435
 436         if (!first_key)
 437                 return 0;
 438
 439         /*
 440          * For live tree block (new tree blocks in current transaction),
 441          * we need proper lock context to avoid race, which is impossible here.
 442          * So we only checks tree blocks which is read from disk, whose
 443          * generation <= fs_info->last_trans_committed.
 444          */
 445         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
 446                 return 0;
 447         if (found_level)
 448                 btrfs_node_key_to_cpu(eb, &found_key, 0);
 449         else
 450                 btrfs_item_key_to_cpu(eb, &found_key, 0);
 451         ret = btrfs_comp_cpu_keys(first_key, &found_key);
 452
 453 #ifdef CONFIG_BTRFS_DEBUG
 454         if (ret) {
 455                 WARN_ON(1);
 456                 btrfs_err(fs_info,
 457 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
 458                           eb->start, parent_transid, first_key->objectid,
 459                           first_key->type, first_key->offset,
 460                           found_key.objectid, found_key.type,
 461                           found_key.offset);
 462         }
 463 #endif
 464         return ret;
 465 }
 466
 467 /*
 468  * helper to read a given tree block, doing retries as required when
 469  * the checksums don't match and we have alternate mirrors to try.
 470  *
 471  * @parent_transid:     expected transid, skip check if 0
 472  * @level:              expected level, mandatory check
 473  * @first_key:          expected key of first slot, skip check if NULL
 474  */
 475 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
 476                                           struct extent_buffer *eb,
 477                                           u64 parent_transid, int level,
 478                                           struct btrfs_key *first_key)
 479 {
 480         struct extent_io_tree *io_tree;
 481         int failed = 0;
 482         int ret;
 483         int num_copies = 0;
 484         int mirror_num = 0;
 485         int failed_mirror = 0;
 486
 487         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
 488         while (1) {
 489                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 490                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
 491                                                mirror_num);
 492                 if (!ret) {
 493                         if (verify_parent_transid(io_tree, eb,
 494                                                    parent_transid, 0))
 495                                 ret = -EIO;
 496                         else if (verify_level_key(fs_info, eb, level,
 497                                                   first_key, parent_transid))
 498                                 ret = -EUCLEAN;
 499                         else
 500                                 break;
 501                 }
 502
 503                 num_copies = btrfs_num_copies(fs_info,
 504                                               eb->start, eb->len);
 505                 if (num_copies == 1)
 506                         break;
 507
 508                 if (!failed_mirror) {
 509                         failed = 1;
 510                         failed_mirror = eb->read_mirror;
 511                 }
 512
 513                 mirror_num++;
 514                 if (mirror_num == failed_mirror)
 515                         mirror_num++;
 516
 517                 if (mirror_num > num_copies)
 518                         break;
 519         }
 520
 521         if (failed && !ret && failed_mirror)
 522                 repair_eb_io_failure(fs_info, eb, failed_mirror);
 523
 524         return ret;
 525 }
 526
 527 /*
 528  * checksum a dirty tree block before IO.  This has extra checks to make sure
 529  * we only fill in the checksum field in the first page of a multi-page block
 530  */
 531
 532 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
 533 {
 534         u64 start = page_offset(page);
 535         u64 found_start;
 536         struct extent_buffer *eb;
 537
 538         eb = (struct extent_buffer *)page->private;
 539         if (page != eb->pages[0])
 540                 return 0;
 541
 542         found_start = btrfs_header_bytenr(eb);
 543         /*
 544          * Please do not consolidate these warnings into a single if.
 545          * It is useful to know what went wrong.
 546          */
 547         if (WARN_ON(found_start != start))
 548                 return -EUCLEAN;
 549         if (WARN_ON(!PageUptodate(page)))
 550                 return -EUCLEAN;
 551
 552         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
 553                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
 554
 555         return csum_tree_block(fs_info, eb, 0);
 556 }
 557
 558 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
 559                                  struct extent_buffer *eb)
 560 {
 561         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 562         u8 fsid[BTRFS_FSID_SIZE];
 563         int ret = 1;
 564
 565         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 566         while (fs_devices) {
 567                 u8 *metadata_uuid;
 568
 569                 /*
 570                  * Checking the incompat flag is only valid for the current
 571                  * fs. For seed devices it's forbidden to have their uuid
 572                  * changed so reading ->fsid in this case is fine
 573                  */
 574                 if (fs_devices == fs_info->fs_devices &&
 575                     btrfs_fs_incompat(fs_info, METADATA_UUID))
 576                         metadata_uuid = fs_devices->metadata_uuid;
 577                 else
 578                         metadata_uuid = fs_devices->fsid;
 579
 580                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
 581                         ret = 0;
 582                         break;
 583                 }
 584                 fs_devices = fs_devices->seed;
 585         }
 586         return ret;
 587 }
 588
 589 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 590                                       u64 phy_offset, struct page *page,
 591                                       u64 start, u64 end, int mirror)
 592 {
 593         u64 found_start;
 594         int found_level;
 595         struct extent_buffer *eb;
 596         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 597         struct btrfs_fs_info *fs_info = root->fs_info;
 598         int ret = 0;
 599         int reads_done;
 600
 601         if (!page->private)
 602                 goto out;
 603
 604         eb = (struct extent_buffer *)page->private;
 605
 606         /* the pending IO might have been the only thing that kept this buffer
 607          * in memory.  Make sure we have a ref for all this other checks
 608          */
 609         extent_buffer_get(eb);
 610
 611         reads_done = atomic_dec_and_test(&eb->io_pages);
 612         if (!reads_done)
 613                 goto err;
 614
 615         eb->read_mirror = mirror;
 616         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
 617                 ret = -EIO;
 618                 goto err;
 619         }
 620
 621         found_start = btrfs_header_bytenr(eb);
 622         if (found_start != eb->start) {
 623                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
 624                              eb->start, found_start);
 625                 ret = -EIO;
 626                 goto err;
 627         }
 628         if (check_tree_block_fsid(fs_info, eb)) {
 629                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
 630                              eb->start);
 631                 ret = -EIO;
 632                 goto err;
 633         }
 634         found_level = btrfs_header_level(eb);
 635         if (found_level >= BTRFS_MAX_LEVEL) {
 636                 btrfs_err(fs_info, "bad tree block level %d on %llu",
 637                           (int)btrfs_header_level(eb), eb->start);
 638                 ret = -EIO;
 639                 goto err;
 640         }
 641
 642         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 643                                        eb, found_level);
 644
 645         ret = csum_tree_block(fs_info, eb, 1);
 646         if (ret)
 647                 goto err;
 648
 649         /*
 650          * If this is a leaf block and it is corrupt, set the corrupt bit so
 651          * that we don't try and read the other copies of this block, just
 652          * return -EIO.
 653          */
 654         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
 655                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 656                 ret = -EIO;
 657         }
 658
 659         if (found_level > 0 && btrfs_check_node(fs_info, eb))
 660                 ret = -EIO;
 661
 662         if (!ret)
 663                 set_extent_buffer_uptodate(eb);
 664 err:
 665         if (reads_done &&
 666             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 667                 btree_readahead_hook(eb, ret);
 668
 669         if (ret) {
 670                 /*
 671                  * our io error hook is going to dec the io pages
 672                  * again, we have to make sure it has something
 673                  * to decrement
 674                  */
 675                 atomic_inc(&eb->io_pages);
 676                 clear_extent_buffer_uptodate(eb);
 677         }
 678         free_extent_buffer(eb);
 679 out:
 680         return ret;
 681 }
 682
 683 static void end_workqueue_bio(struct bio *bio)
 684 {
 685         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
 686         struct btrfs_fs_info *fs_info;
 687         struct btrfs_workqueue *wq;
 688         btrfs_work_func_t func;
 689
 690         fs_info = end_io_wq->info;
 691         end_io_wq->status = bio->bi_status;
 692
 693         if (bio_op(bio) == REQ_OP_WRITE) {
 694                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
 695                         wq = fs_info->endio_meta_write_workers;
 696                         func = btrfs_endio_meta_write_helper;
 697                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
 698                         wq = fs_info->endio_freespace_worker;
 699                         func = btrfs_freespace_write_helper;
 700                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 701                         wq = fs_info->endio_raid56_workers;
 702                         func = btrfs_endio_raid56_helper;
 703                 } else {
 704                         wq = fs_info->endio_write_workers;
 705                         func = btrfs_endio_write_helper;
 706                 }
 707         } else {
 708                 if (unlikely(end_io_wq->metadata ==
 709                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
 710                         wq = fs_info->endio_repair_workers;
 711                         func = btrfs_endio_repair_helper;
 712                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
 713                         wq = fs_info->endio_raid56_workers;
 714                         func = btrfs_endio_raid56_helper;
 715                 } else if (end_io_wq->metadata) {
 716                         wq = fs_info->endio_meta_workers;
 717                         func = btrfs_endio_meta_helper;
 718                 } else {
 719                         wq = fs_info->endio_workers;
 720                         func = btrfs_endio_helper;
 721                 }
 722         }
 723
 724         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
 725         btrfs_queue_work(wq, &end_io_wq->work);
 726 }
 727
 728 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 729                         enum btrfs_wq_endio_type metadata)
 730 {
 731         struct btrfs_end_io_wq *end_io_wq;
 732
 733         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
 734         if (!end_io_wq)
 735                 return BLK_STS_RESOURCE;
 736
 737         end_io_wq->private = bio->bi_private;
 738         end_io_wq->end_io = bio->bi_end_io;
 739         end_io_wq->info = info;
 740         end_io_wq->status = 0;
 741         end_io_wq->bio = bio;
 742         end_io_wq->metadata = metadata;
 743
 744         bio->bi_private = end_io_wq;
 745         bio->bi_end_io = end_workqueue_bio;
 746         return 0;
 747 }
 748
 749 static void run_one_async_start(struct btrfs_work *work)
 750 {
 751         struct async_submit_bio *async;
 752         blk_status_t ret;
 753
 754         async = container_of(work, struct  async_submit_bio, work);
 755         ret = async->submit_bio_start(async->private_data, async->bio,
 756                                       async->bio_offset);
 757         if (ret)
 758                 async->status = ret;
 759 }
 760
 761 /*
 762  * In order to insert checksums into the metadata in large chunks, we wait
 763  * until bio submission time.   All the pages in the bio are checksummed and
 764  * sums are attached onto the ordered extent record.
 765  *
 766  * At IO completion time the csums attached on the ordered extent record are
 767  * inserted into the tree.
 768  */
 769 static void run_one_async_done(struct btrfs_work *work)
 770 {
 771         struct async_submit_bio *async;
 772         struct inode *inode;
 773         blk_status_t ret;
 774
 775         async = container_of(work, struct  async_submit_bio, work);
 776         inode = async->private_data;
 777
 778         /* If an error occurred we just want to clean up the bio and move on */
 779         if (async->status) {
 780                 async->bio->bi_status = async->status;
 781                 bio_endio(async->bio);
 782                 return;
 783         }
 784
 785         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
 786                         async->mirror_num, 1);
 787         if (ret) {
 788                 async->bio->bi_status = ret;
 789                 bio_endio(async->bio);
 790         }
 791 }
 792
 793 static void run_one_async_free(struct btrfs_work *work)
 794 {
 795         struct async_submit_bio *async;
 796
 797         async = container_of(work, struct  async_submit_bio, work);
 798         kfree(async);
 799 }
 800
 801 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
 802                                  int mirror_num, unsigned long bio_flags,
 803                                  u64 bio_offset, void *private_data,
 804                                  extent_submit_bio_start_t *submit_bio_start)
 805 {
 806         struct async_submit_bio *async;
 807
 808         async = kmalloc(sizeof(*async), GFP_NOFS);
 809         if (!async)
 810                 return BLK_STS_RESOURCE;
 811
 812         async->private_data = private_data;
 813         async->bio = bio;
 814         async->mirror_num = mirror_num;
 815         async->submit_bio_start = submit_bio_start;
 816
 817         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
 818                         run_one_async_done, run_one_async_free);
 819
 820         async->bio_offset = bio_offset;
 821
 822         async->status = 0;
 823
 824         if (op_is_sync(bio->bi_opf))
 825                 btrfs_set_work_high_priority(&async->work);
 826
 827         btrfs_queue_work(fs_info->workers, &async->work);
 828         return 0;
 829 }
 830
 831 static blk_status_t btree_csum_one_bio(struct bio *bio)
 832 {
 833         struct bio_vec *bvec;
 834         struct btrfs_root *root;
 835         int i, ret = 0;
 836         struct bvec_iter_all iter_all;
 837
 838         ASSERT(!bio_flagged(bio, BIO_CLONED));
 839         bio_for_each_segment_all(bvec, bio, i, iter_all) {
 840                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 841                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
 842                 if (ret)
 843                         break;
 844         }
 845
 846         return errno_to_blk_status(ret);
 847 }
 848
 849 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
 850                                              u64 bio_offset)
 851 {
 852         /*
 853          * when we're called for a write, we're already in the async
 854          * submission context.  Just jump into btrfs_map_bio
 855          */
 856         return btree_csum_one_bio(bio);
 857 }
 858
 859 static int check_async_write(struct btrfs_inode *bi)
 860 {
 861         if (atomic_read(&bi->sync_writers))
 862                 return 0;
 863 #ifdef CONFIG_X86
 864         if (static_cpu_has(X86_FEATURE_XMM4_2))
 865                 return 0;
 866 #endif
 867         return 1;
 868 }
 869
 870 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
 871                                           int mirror_num, unsigned long bio_flags,
 872                                           u64 bio_offset)
 873 {
 874         struct inode *inode = private_data;
 875         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 876         int async = check_async_write(BTRFS_I(inode));
 877         blk_status_t ret;
 878
 879         if (bio_op(bio) != REQ_OP_WRITE) {
 880                 /*
 881                  * called for a read, do the setup so that checksum validation
 882                  * can happen in the async kernel threads
 883                  */
 884                 ret = btrfs_bio_wq_end_io(fs_info, bio,
 885                                           BTRFS_WQ_ENDIO_METADATA);
 886                 if (ret)
 887                         goto out_w_error;
 888                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 889         } else if (!async) {
 890                 ret = btree_csum_one_bio(bio);
 891                 if (ret)
 892                         goto out_w_error;
 893                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
 894         } else {
 895                 /*
 896                  * kthread helpers are used to submit writes so that
 897                  * checksumming can happen in parallel across all CPUs
 898                  */
 899                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
 900                                           bio_offset, private_data,
 901                                           btree_submit_bio_start);
 902         }
 903
 904         if (ret)
 905                 goto out_w_error;
 906         return 0;
 907
 908 out_w_error:
 909         bio->bi_status = ret;
 910         bio_endio(bio);
 911         return ret;
 912 }
 913
 914 #ifdef CONFIG_MIGRATION
 915 static int btree_migratepage(struct address_space *mapping,
 916                         struct page *newpage, struct page *page,
 917                         enum migrate_mode mode)
 918 {
 919         /*
 920          * we can't safely write a btree page from here,
 921          * we haven't done the locking hook
 922          */
 923         if (PageDirty(page))
 924                 return -EAGAIN;
 925         /*
 926          * Buffers may be managed in a filesystem specific way.
 927          * We must have no buffers or drop them.
 928          */
 929         if (page_has_private(page) &&
 930             !try_to_release_page(page, GFP_KERNEL))
 931                 return -EAGAIN;
 932         return migrate_page(mapping, newpage, page, mode);
 933 }
 934 #endif
 935
 936
 937 static int btree_writepages(struct address_space *mapping,
 938                             struct writeback_control *wbc)
 939 {
 940         struct btrfs_fs_info *fs_info;
 941         int ret;
 942
 943         if (wbc->sync_mode == WB_SYNC_NONE) {
 944
 945                 if (wbc->for_kupdate)
 946                         return 0;
 947
 948                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 949                 /* this is a bit racy, but that's ok */
 950                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 951                                              BTRFS_DIRTY_METADATA_THRESH,
 952                                              fs_info->dirty_metadata_batch);
 953                 if (ret < 0)
 954                         return 0;
 955         }
 956         return btree_write_cache_pages(mapping, wbc);
 957 }
 958
 959 static int btree_readpage(struct file *file, struct page *page)
 960 {
 961         struct extent_io_tree *tree;
 962         tree = &BTRFS_I(page->mapping->host)->io_tree;
 963         return extent_read_full_page(tree, page, btree_get_extent, 0);
 964 }
 965
 966 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 967 {
 968         if (PageWriteback(page) || PageDirty(page))
 969                 return 0;
 970
 971         return try_release_extent_buffer(page);
 972 }
 973
 974 static void btree_invalidatepage(struct page *page, unsigned int offset,
 975                                  unsigned int length)
 976 {
 977         struct extent_io_tree *tree;
 978         tree = &BTRFS_I(page->mapping->host)->io_tree;
 979         extent_invalidatepage(tree, page, offset);
 980         btree_releasepage(page, GFP_NOFS);
 981         if (PagePrivate(page)) {
 982                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
 983                            "page private not zero on page %llu",
 984                            (unsigned long long)page_offset(page));
 985                 ClearPagePrivate(page);
 986                 set_page_private(page, 0);
 987                 put_page(page);
 988         }
 989 }
 990
 991 static int btree_set_page_dirty(struct page *page)
 992 {
 993 #ifdef DEBUG
 994         struct extent_buffer *eb;
 995
 996         BUG_ON(!PagePrivate(page));
 997         eb = (struct extent_buffer *)page->private;
 998         BUG_ON(!eb);
 999         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1000         BUG_ON(!atomic_read(&eb->refs));
1001         btrfs_assert_tree_locked(eb);
1002 #endif
1003         return __set_page_dirty_nobuffers(page);
1004 }
1005
1006 static const struct address_space_operations btree_aops = {
1007         .readpage       = btree_readpage,
1008         .writepages     = btree_writepages,
1009         .releasepage    = btree_releasepage,
1010         .invalidatepage = btree_invalidatepage,
1011 #ifdef CONFIG_MIGRATION
1012         .migratepage    = btree_migratepage,
1013 #endif
1014         .set_page_dirty = btree_set_page_dirty,
1015 };
1016
1017 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1018 {
1019         struct extent_buffer *buf = NULL;
1020         struct inode *btree_inode = fs_info->btree_inode;
1021
1022         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1023         if (IS_ERR(buf))
1024                 return;
1025         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1026                                  buf, WAIT_NONE, 0);
1027         free_extent_buffer(buf);
1028 }
1029
1030 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1031                          int mirror_num, struct extent_buffer **eb)
1032 {
1033         struct extent_buffer *buf = NULL;
1034         struct inode *btree_inode = fs_info->btree_inode;
1035         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1036         int ret;
1037
1038         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1039         if (IS_ERR(buf))
1040                 return 0;
1041
1042         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1043
1044         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1045                                        mirror_num);
1046         if (ret) {
1047                 free_extent_buffer(buf);
1048                 return ret;
1049         }
1050
1051         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1052                 free_extent_buffer(buf);
1053                 return -EIO;
1054         } else if (extent_buffer_uptodate(buf)) {
1055                 *eb = buf;
1056         } else {
1057                 free_extent_buffer(buf);
1058         }
1059         return 0;
1060 }
1061
1062 struct extent_buffer *btrfs_find_create_tree_block(
1063                                                 struct btrfs_fs_info *fs_info,
1064                                                 u64 bytenr)
1065 {
1066         if (btrfs_is_testing(fs_info))
1067                 return alloc_test_extent_buffer(fs_info, bytenr);
1068         return alloc_extent_buffer(fs_info, bytenr);
1069 }
1070
1071
1072 int btrfs_write_tree_block(struct extent_buffer *buf)
1073 {
1074         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1075                                         buf->start + buf->len - 1);
1076 }
1077
1078 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1079 {
1080         filemap_fdatawait_range(buf->pages[0]->mapping,
1081                                 buf->start, buf->start + buf->len - 1);
1082 }
1083
1084 /*
1085  * Read tree block at logical address @bytenr and do variant basic but critical
1086  * verification.
1087  *
1088  * @parent_transid:     expected transid of this tree block, skip check if 0
1089  * @level:              expected level, mandatory check
1090  * @first_key:          expected key in slot 0, skip check if NULL
1091  */
1092 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1093                                       u64 parent_transid, int level,
1094                                       struct btrfs_key *first_key)
1095 {
1096         struct extent_buffer *buf = NULL;
1097         int ret;
1098
1099         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1100         if (IS_ERR(buf))
1101                 return buf;
1102
1103         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1104                                              level, first_key);
1105         if (ret) {
1106                 free_extent_buffer(buf);
1107                 return ERR_PTR(ret);
1108         }
1109         return buf;
1110
1111 }
1112
1113 void clean_tree_block(struct btrfs_fs_info *fs_info,
1114                       struct extent_buffer *buf)
1115 {
1116         if (btrfs_header_generation(buf) ==
1117             fs_info->running_transaction->transid) {
1118                 btrfs_assert_tree_locked(buf);
1119
1120                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1121                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1122                                                  -buf->len,
1123                                                  fs_info->dirty_metadata_batch);
1124                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1125                         btrfs_set_lock_blocking_write(buf);
1126                         clear_extent_buffer_dirty(buf);
1127                 }
1128         }
1129 }
1130
1131 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1132 {
1133         struct btrfs_subvolume_writers *writers;
1134         int ret;
1135
1136         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1137         if (!writers)
1138                 return ERR_PTR(-ENOMEM);
1139
1140         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1141         if (ret < 0) {
1142                 kfree(writers);
1143                 return ERR_PTR(ret);
1144         }
1145
1146         init_waitqueue_head(&writers->wait);
1147         return writers;
1148 }
1149
1150 static void
1151 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1152 {
1153         percpu_counter_destroy(&writers->counter);
1154         kfree(writers);
1155 }
1156
1157 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1158                          u64 objectid)
1159 {
1160         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1161         root->node = NULL;
1162         root->commit_root = NULL;
1163         root->state = 0;
1164         root->orphan_cleanup_state = 0;
1165
1166         root->last_trans = 0;
1167         root->highest_objectid = 0;
1168         root->nr_delalloc_inodes = 0;
1169         root->nr_ordered_extents = 0;
1170         root->inode_tree = RB_ROOT;
1171         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1172         root->block_rsv = NULL;
1173
1174         INIT_LIST_HEAD(&root->dirty_list);
1175         INIT_LIST_HEAD(&root->root_list);
1176         INIT_LIST_HEAD(&root->delalloc_inodes);
1177         INIT_LIST_HEAD(&root->delalloc_root);
1178         INIT_LIST_HEAD(&root->ordered_extents);
1179         INIT_LIST_HEAD(&root->ordered_root);
1180         INIT_LIST_HEAD(&root->reloc_dirty_list);
1181         INIT_LIST_HEAD(&root->logged_list[0]);
1182         INIT_LIST_HEAD(&root->logged_list[1]);
1183         spin_lock_init(&root->inode_lock);
1184         spin_lock_init(&root->delalloc_lock);
1185         spin_lock_init(&root->ordered_extent_lock);
1186         spin_lock_init(&root->accounting_lock);
1187         spin_lock_init(&root->log_extents_lock[0]);
1188         spin_lock_init(&root->log_extents_lock[1]);
1189         spin_lock_init(&root->qgroup_meta_rsv_lock);
1190         mutex_init(&root->objectid_mutex);
1191         mutex_init(&root->log_mutex);
1192         mutex_init(&root->ordered_extent_mutex);
1193         mutex_init(&root->delalloc_mutex);
1194         init_waitqueue_head(&root->log_writer_wait);
1195         init_waitqueue_head(&root->log_commit_wait[0]);
1196         init_waitqueue_head(&root->log_commit_wait[1]);
1197         INIT_LIST_HEAD(&root->log_ctxs[0]);
1198         INIT_LIST_HEAD(&root->log_ctxs[1]);
1199         atomic_set(&root->log_commit[0], 0);
1200         atomic_set(&root->log_commit[1], 0);
1201         atomic_set(&root->log_writers, 0);
1202         atomic_set(&root->log_batch, 0);
1203         refcount_set(&root->refs, 1);
1204         atomic_set(&root->will_be_snapshotted, 0);
1205         atomic_set(&root->snapshot_force_cow, 0);
1206         atomic_set(&root->nr_swapfiles, 0);
1207         root->log_transid = 0;
1208         root->log_transid_committed = -1;
1209         root->last_log_commit = 0;
1210         if (!dummy)
1211                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1212
1213         memset(&root->root_key, 0, sizeof(root->root_key));
1214         memset(&root->root_item, 0, sizeof(root->root_item));
1215         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1216         if (!dummy)
1217                 root->defrag_trans_start = fs_info->generation;
1218         else
1219                 root->defrag_trans_start = 0;
1220         root->root_key.objectid = objectid;
1221         root->anon_dev = 0;
1222
1223         spin_lock_init(&root->root_item_lock);
1224         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1225 }
1226
1227 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1228                 gfp_t flags)
1229 {
1230         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1231         if (root)
1232                 root->fs_info = fs_info;
1233         return root;
1234 }
1235
1236 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1237 /* Should only be used by the testing infrastructure */
1238 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1239 {
1240         struct btrfs_root *root;
1241
1242         if (!fs_info)
1243                 return ERR_PTR(-EINVAL);
1244
1245         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1246         if (!root)
1247                 return ERR_PTR(-ENOMEM);
1248
1249         /* We don't use the stripesize in selftest, set it as sectorsize */
1250         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1251         root->alloc_bytenr = 0;
1252
1253         return root;
1254 }
1255 #endif
1256
1257 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1258                                      struct btrfs_fs_info *fs_info,
1259                                      u64 objectid)
1260 {
1261         struct extent_buffer *leaf;
1262         struct btrfs_root *tree_root = fs_info->tree_root;
1263         struct btrfs_root *root;
1264         struct btrfs_key key;
1265         unsigned int nofs_flag;
1266         int ret = 0;
1267         uuid_le uuid = NULL_UUID_LE;
1268
1269         /*
1270          * We're holding a transaction handle, so use a NOFS memory allocation
1271          * context to avoid deadlock if reclaim happens.
1272          */
1273         nofs_flag = memalloc_nofs_save();
1274         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1275         memalloc_nofs_restore(nofs_flag);
1276         if (!root)
1277                 return ERR_PTR(-ENOMEM);
1278
1279         __setup_root(root, fs_info, objectid);
1280         root->root_key.objectid = objectid;
1281         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1282         root->root_key.offset = 0;
1283
1284         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1285         if (IS_ERR(leaf)) {
1286                 ret = PTR_ERR(leaf);
1287                 leaf = NULL;
1288                 goto fail;
1289         }
1290
1291         root->node = leaf;
1292         btrfs_mark_buffer_dirty(leaf);
1293
1294         root->commit_root = btrfs_root_node(root);
1295         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1296
1297         root->root_item.flags = 0;
1298         root->root_item.byte_limit = 0;
1299         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1300         btrfs_set_root_generation(&root->root_item, trans->transid);
1301         btrfs_set_root_level(&root->root_item, 0);
1302         btrfs_set_root_refs(&root->root_item, 1);
1303         btrfs_set_root_used(&root->root_item, leaf->len);
1304         btrfs_set_root_last_snapshot(&root->root_item, 0);
1305         btrfs_set_root_dirid(&root->root_item, 0);
1306         if (is_fstree(objectid))
1307                 uuid_le_gen(&uuid);
1308         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1309         root->root_item.drop_level = 0;
1310
1311         key.objectid = objectid;
1312         key.type = BTRFS_ROOT_ITEM_KEY;
1313         key.offset = 0;
1314         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1315         if (ret)
1316                 goto fail;
1317
1318         btrfs_tree_unlock(leaf);
1319
1320         return root;
1321
1322 fail:
1323         if (leaf) {
1324                 btrfs_tree_unlock(leaf);
1325                 free_extent_buffer(root->commit_root);
1326                 free_extent_buffer(leaf);
1327         }
1328         kfree(root);
1329
1330         return ERR_PTR(ret);
1331 }
1332
1333 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1334                                          struct btrfs_fs_info *fs_info)
1335 {
1336         struct btrfs_root *root;
1337         struct extent_buffer *leaf;
1338
1339         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1340         if (!root)
1341                 return ERR_PTR(-ENOMEM);
1342
1343         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1344
1345         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1346         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1347         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1348
1349         /*
1350          * DON'T set REF_COWS for log trees
1351          *
1352          * log trees do not get reference counted because they go away
1353          * before a real commit is actually done.  They do store pointers
1354          * to file data extents, and those reference counts still get
1355          * updated (along with back refs to the log tree).
1356          */
1357
1358         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1359                         NULL, 0, 0, 0);
1360         if (IS_ERR(leaf)) {
1361                 kfree(root);
1362                 return ERR_CAST(leaf);
1363         }
1364
1365         root->node = leaf;
1366
1367         btrfs_mark_buffer_dirty(root->node);
1368         btrfs_tree_unlock(root->node);
1369         return root;
1370 }
1371
1372 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1373                              struct btrfs_fs_info *fs_info)
1374 {
1375         struct btrfs_root *log_root;
1376
1377         log_root = alloc_log_tree(trans, fs_info);
1378         if (IS_ERR(log_root))
1379                 return PTR_ERR(log_root);
1380         WARN_ON(fs_info->log_root_tree);
1381         fs_info->log_root_tree = log_root;
1382         return 0;
1383 }
1384
1385 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1386                        struct btrfs_root *root)
1387 {
1388         struct btrfs_fs_info *fs_info = root->fs_info;
1389         struct btrfs_root *log_root;
1390         struct btrfs_inode_item *inode_item;
1391
1392         log_root = alloc_log_tree(trans, fs_info);
1393         if (IS_ERR(log_root))
1394                 return PTR_ERR(log_root);
1395
1396         log_root->last_trans = trans->transid;
1397         log_root->root_key.offset = root->root_key.objectid;
1398
1399         inode_item = &log_root->root_item.inode;
1400         btrfs_set_stack_inode_generation(inode_item, 1);
1401         btrfs_set_stack_inode_size(inode_item, 3);
1402         btrfs_set_stack_inode_nlink(inode_item, 1);
1403         btrfs_set_stack_inode_nbytes(inode_item,
1404                                      fs_info->nodesize);
1405         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1406
1407         btrfs_set_root_node(&log_root->root_item, log_root->node);
1408
1409         WARN_ON(root->log_root);
1410         root->log_root = log_root;
1411         root->log_transid = 0;
1412         root->log_transid_committed = -1;
1413         root->last_log_commit = 0;
1414         return 0;
1415 }
1416
1417 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1418                                                struct btrfs_key *key)
1419 {
1420         struct btrfs_root *root;
1421         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1422         struct btrfs_path *path;
1423         u64 generation;
1424         int ret;
1425         int level;
1426
1427         path = btrfs_alloc_path();
1428         if (!path)
1429                 return ERR_PTR(-ENOMEM);
1430
1431         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1432         if (!root) {
1433                 ret = -ENOMEM;
1434                 goto alloc_fail;
1435         }
1436
1437         __setup_root(root, fs_info, key->objectid);
1438
1439         ret = btrfs_find_root(tree_root, key, path,
1440                               &root->root_item, &root->root_key);
1441         if (ret) {
1442                 if (ret > 0)
1443                         ret = -ENOENT;
1444                 goto find_fail;
1445         }
1446
1447         generation = btrfs_root_generation(&root->root_item);
1448         level = btrfs_root_level(&root->root_item);
1449         root->node = read_tree_block(fs_info,
1450                                      btrfs_root_bytenr(&root->root_item),
1451                                      generation, level, NULL);
1452         if (IS_ERR(root->node)) {
1453                 ret = PTR_ERR(root->node);
1454                 goto find_fail;
1455         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1456                 ret = -EIO;
1457                 free_extent_buffer(root->node);
1458                 goto find_fail;
1459         }
1460         root->commit_root = btrfs_root_node(root);
1461 out:
1462         btrfs_free_path(path);
1463         return root;
1464
1465 find_fail:
1466         kfree(root);
1467 alloc_fail:
1468         root = ERR_PTR(ret);
1469         goto out;
1470 }
1471
1472 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1473                                       struct btrfs_key *location)
1474 {
1475         struct btrfs_root *root;
1476
1477         root = btrfs_read_tree_root(tree_root, location);
1478         if (IS_ERR(root))
1479                 return root;
1480
1481         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1482                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1483                 btrfs_check_and_init_root_item(&root->root_item);
1484         }
1485
1486         return root;
1487 }
1488
1489 int btrfs_init_fs_root(struct btrfs_root *root)
1490 {
1491         int ret;
1492         struct btrfs_subvolume_writers *writers;
1493
1494         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1495         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1496                                         GFP_NOFS);
1497         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1498                 ret = -ENOMEM;
1499                 goto fail;
1500         }
1501
1502         writers = btrfs_alloc_subvolume_writers();
1503         if (IS_ERR(writers)) {
1504                 ret = PTR_ERR(writers);
1505                 goto fail;
1506         }
1507         root->subv_writers = writers;
1508
1509         btrfs_init_free_ino_ctl(root);
1510         spin_lock_init(&root->ino_cache_lock);
1511         init_waitqueue_head(&root->ino_cache_wait);
1512
1513         ret = get_anon_bdev(&root->anon_dev);
1514         if (ret)
1515                 goto fail;
1516
1517         mutex_lock(&root->objectid_mutex);
1518         ret = btrfs_find_highest_objectid(root,
1519                                         &root->highest_objectid);
1520         if (ret) {
1521                 mutex_unlock(&root->objectid_mutex);
1522                 goto fail;
1523         }
1524
1525         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1526
1527         mutex_unlock(&root->objectid_mutex);
1528
1529         return 0;
1530 fail:
1531         /* The caller is responsible to call btrfs_free_fs_root */
1532         return ret;
1533 }
1534
1535 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1536                                         u64 root_id)
1537 {
1538         struct btrfs_root *root;
1539
1540         spin_lock(&fs_info->fs_roots_radix_lock);
1541         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1542                                  (unsigned long)root_id);
1543         spin_unlock(&fs_info->fs_roots_radix_lock);
1544         return root;
1545 }
1546
1547 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1548                          struct btrfs_root *root)
1549 {
1550         int ret;
1551
1552         ret = radix_tree_preload(GFP_NOFS);
1553         if (ret)
1554                 return ret;
1555
1556         spin_lock(&fs_info->fs_roots_radix_lock);
1557         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1558                                 (unsigned long)root->root_key.objectid,
1559                                 root);
1560         if (ret == 0)
1561                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1562         spin_unlock(&fs_info->fs_roots_radix_lock);
1563         radix_tree_preload_end();
1564
1565         return ret;
1566 }
1567
1568 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1569                                      struct btrfs_key *location,
1570                                      bool check_ref)
1571 {
1572         struct btrfs_root *root;
1573         struct btrfs_path *path;
1574         struct btrfs_key key;
1575         int ret;
1576
1577         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1578                 return fs_info->tree_root;
1579         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1580                 return fs_info->extent_root;
1581         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1582                 return fs_info->chunk_root;
1583         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1584                 return fs_info->dev_root;
1585         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1586                 return fs_info->csum_root;
1587         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1588                 return fs_info->quota_root ? fs_info->quota_root :
1589                                              ERR_PTR(-ENOENT);
1590         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1591                 return fs_info->uuid_root ? fs_info->uuid_root :
1592                                             ERR_PTR(-ENOENT);
1593         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1594                 return fs_info->free_space_root ? fs_info->free_space_root :
1595                                                   ERR_PTR(-ENOENT);
1596 again:
1597         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1598         if (root) {
1599                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1600                         return ERR_PTR(-ENOENT);
1601                 return root;
1602         }
1603
1604         root = btrfs_read_fs_root(fs_info->tree_root, location);
1605         if (IS_ERR(root))
1606                 return root;
1607
1608         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1609                 ret = -ENOENT;
1610                 goto fail;
1611         }
1612
1613         ret = btrfs_init_fs_root(root);
1614         if (ret)
1615                 goto fail;
1616
1617         path = btrfs_alloc_path();
1618         if (!path) {
1619                 ret = -ENOMEM;
1620                 goto fail;
1621         }
1622         key.objectid = BTRFS_ORPHAN_OBJECTID;
1623         key.type = BTRFS_ORPHAN_ITEM_KEY;
1624         key.offset = location->objectid;
1625
1626         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1627         btrfs_free_path(path);
1628         if (ret < 0)
1629                 goto fail;
1630         if (ret == 0)
1631                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1632
1633         ret = btrfs_insert_fs_root(fs_info, root);
1634         if (ret) {
1635                 if (ret == -EEXIST) {
1636                         btrfs_free_fs_root(root);
1637                         goto again;
1638                 }
1639                 goto fail;
1640         }
1641         return root;
1642 fail:
1643         btrfs_free_fs_root(root);
1644         return ERR_PTR(ret);
1645 }
1646
1647 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1648 {
1649         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1650         int ret = 0;
1651         struct btrfs_device *device;
1652         struct backing_dev_info *bdi;
1653
1654         rcu_read_lock();
1655         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1656                 if (!device->bdev)
1657                         continue;
1658                 bdi = device->bdev->bd_bdi;
1659                 if (bdi_congested(bdi, bdi_bits)) {
1660                         ret = 1;
1661                         break;
1662                 }
1663         }
1664         rcu_read_unlock();
1665         return ret;
1666 }
1667
1668 /*
1669  * called by the kthread helper functions to finally call the bio end_io
1670  * functions.  This is where read checksum verification actually happens
1671  */
1672 static void end_workqueue_fn(struct btrfs_work *work)
1673 {
1674         struct bio *bio;
1675         struct btrfs_end_io_wq *end_io_wq;
1676
1677         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1678         bio = end_io_wq->bio;
1679
1680         bio->bi_status = end_io_wq->status;
1681         bio->bi_private = end_io_wq->private;
1682         bio->bi_end_io = end_io_wq->end_io;
1683         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1684         bio_endio(bio);
1685 }
1686
1687 static int cleaner_kthread(void *arg)
1688 {
1689         struct btrfs_root *root = arg;
1690         struct btrfs_fs_info *fs_info = root->fs_info;
1691         int again;
1692
1693         while (1) {
1694                 again = 0;
1695
1696                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1697
1698                 /* Make the cleaner go to sleep early. */
1699                 if (btrfs_need_cleaner_sleep(fs_info))
1700                         goto sleep;
1701
1702                 /*
1703                  * Do not do anything if we might cause open_ctree() to block
1704                  * before we have finished mounting the filesystem.
1705                  */
1706                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1707                         goto sleep;
1708
1709                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1710                         goto sleep;
1711
1712                 /*
1713                  * Avoid the problem that we change the status of the fs
1714                  * during the above check and trylock.
1715                  */
1716                 if (btrfs_need_cleaner_sleep(fs_info)) {
1717                         mutex_unlock(&fs_info->cleaner_mutex);
1718                         goto sleep;
1719                 }
1720
1721                 btrfs_run_delayed_iputs(fs_info);
1722
1723                 again = btrfs_clean_one_deleted_snapshot(root);
1724                 mutex_unlock(&fs_info->cleaner_mutex);
1725
1726                 /*
1727                  * The defragger has dealt with the R/O remount and umount,
1728                  * needn't do anything special here.
1729                  */
1730                 btrfs_run_defrag_inodes(fs_info);
1731
1732                 /*
1733                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1734                  * with relocation (btrfs_relocate_chunk) and relocation
1735                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1736                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1737                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1738                  * unused block groups.
1739                  */
1740                 btrfs_delete_unused_bgs(fs_info);
1741 sleep:
1742                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1743                 if (kthread_should_park())
1744                         kthread_parkme();
1745                 if (kthread_should_stop())
1746                         return 0;
1747                 if (!again) {
1748                         set_current_state(TASK_INTERRUPTIBLE);
1749                         schedule();
1750                         __set_current_state(TASK_RUNNING);
1751                 }
1752         }
1753 }
1754
1755 static int transaction_kthread(void *arg)
1756 {
1757         struct btrfs_root *root = arg;
1758         struct btrfs_fs_info *fs_info = root->fs_info;
1759         struct btrfs_trans_handle *trans;
1760         struct btrfs_transaction *cur;
1761         u64 transid;
1762         time64_t now;
1763         unsigned long delay;
1764         bool cannot_commit;
1765
1766         do {
1767                 cannot_commit = false;
1768                 delay = HZ * fs_info->commit_interval;
1769                 mutex_lock(&fs_info->transaction_kthread_mutex);
1770
1771                 spin_lock(&fs_info->trans_lock);
1772                 cur = fs_info->running_transaction;
1773                 if (!cur) {
1774                         spin_unlock(&fs_info->trans_lock);
1775                         goto sleep;
1776                 }
1777
1778                 now = ktime_get_seconds();
1779                 if (cur->state < TRANS_STATE_BLOCKED &&
1780                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1781                     (now < cur->start_time ||
1782                      now - cur->start_time < fs_info->commit_interval)) {
1783                         spin_unlock(&fs_info->trans_lock);
1784                         delay = HZ * 5;
1785                         goto sleep;
1786                 }
1787                 transid = cur->transid;
1788                 spin_unlock(&fs_info->trans_lock);
1789
1790                 /* If the file system is aborted, this will always fail. */
1791                 trans = btrfs_attach_transaction(root);
1792                 if (IS_ERR(trans)) {
1793                         if (PTR_ERR(trans) != -ENOENT)
1794                                 cannot_commit = true;
1795                         goto sleep;
1796                 }
1797                 if (transid == trans->transid) {
1798                         btrfs_commit_transaction(trans);
1799                 } else {
1800                         btrfs_end_transaction(trans);
1801                 }
1802 sleep:
1803                 wake_up_process(fs_info->cleaner_kthread);
1804                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1805
1806                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1807                                       &fs_info->fs_state)))
1808                         btrfs_cleanup_transaction(fs_info);
1809                 if (!kthread_should_stop() &&
1810                                 (!btrfs_transaction_blocked(fs_info) ||
1811                                  cannot_commit))
1812                         schedule_timeout_interruptible(delay);
1813         } while (!kthread_should_stop());
1814         return 0;
1815 }
1816
1817 /*
1818  * this will find the highest generation in the array of
1819  * root backups.  The index of the highest array is returned,
1820  * or -1 if we can't find anything.
1821  *
1822  * We check to make sure the array is valid by comparing the
1823  * generation of the latest  root in the array with the generation
1824  * in the super block.  If they don't match we pitch it.
1825  */
1826 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1827 {
1828         u64 cur;
1829         int newest_index = -1;
1830         struct btrfs_root_backup *root_backup;
1831         int i;
1832
1833         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1834                 root_backup = info->super_copy->super_roots + i;
1835                 cur = btrfs_backup_tree_root_gen(root_backup);
1836                 if (cur == newest_gen)
1837                         newest_index = i;
1838         }
1839
1840         /* check to see if we actually wrapped around */
1841         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1842                 root_backup = info->super_copy->super_roots;
1843                 cur = btrfs_backup_tree_root_gen(root_backup);
1844                 if (cur == newest_gen)
1845                         newest_index = 0;
1846         }
1847         return newest_index;
1848 }
1849
1850
1851 /*
1852  * find the oldest backup so we know where to store new entries
1853  * in the backup array.  This will set the backup_root_index
1854  * field in the fs_info struct
1855  */
1856 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1857                                      u64 newest_gen)
1858 {
1859         int newest_index = -1;
1860
1861         newest_index = find_newest_super_backup(info, newest_gen);
1862         /* if there was garbage in there, just move along */
1863         if (newest_index == -1) {
1864                 info->backup_root_index = 0;
1865         } else {
1866                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1867         }
1868 }
1869
1870 /*
1871  * copy all the root pointers into the super backup array.
1872  * this will bump the backup pointer by one when it is
1873  * done
1874  */
1875 static void backup_super_roots(struct btrfs_fs_info *info)
1876 {
1877         int next_backup;
1878         struct btrfs_root_backup *root_backup;
1879         int last_backup;
1880
1881         next_backup = info->backup_root_index;
1882         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1883                 BTRFS_NUM_BACKUP_ROOTS;
1884
1885         /*
1886          * just overwrite the last backup if we're at the same generation
1887          * this happens only at umount
1888          */
1889         root_backup = info->super_for_commit->super_roots + last_backup;
1890         if (btrfs_backup_tree_root_gen(root_backup) ==
1891             btrfs_header_generation(info->tree_root->node))
1892                 next_backup = last_backup;
1893
1894         root_backup = info->super_for_commit->super_roots + next_backup;
1895
1896         /*
1897          * make sure all of our padding and empty slots get zero filled
1898          * regardless of which ones we use today
1899          */
1900         memset(root_backup, 0, sizeof(*root_backup));
1901
1902         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1903
1904         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1905         btrfs_set_backup_tree_root_gen(root_backup,
1906                                btrfs_header_generation(info->tree_root->node));
1907
1908         btrfs_set_backup_tree_root_level(root_backup,
1909                                btrfs_header_level(info->tree_root->node));
1910
1911         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1912         btrfs_set_backup_chunk_root_gen(root_backup,
1913                                btrfs_header_generation(info->chunk_root->node));
1914         btrfs_set_backup_chunk_root_level(root_backup,
1915                                btrfs_header_level(info->chunk_root->node));
1916
1917         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1918         btrfs_set_backup_extent_root_gen(root_backup,
1919                                btrfs_header_generation(info->extent_root->node));
1920         btrfs_set_backup_extent_root_level(root_backup,
1921                                btrfs_header_level(info->extent_root->node));
1922
1923         /*
1924          * we might commit during log recovery, which happens before we set
1925          * the fs_root.  Make sure it is valid before we fill it in.
1926          */
1927         if (info->fs_root && info->fs_root->node) {
1928                 btrfs_set_backup_fs_root(root_backup,
1929                                          info->fs_root->node->start);
1930                 btrfs_set_backup_fs_root_gen(root_backup,
1931                                btrfs_header_generation(info->fs_root->node));
1932                 btrfs_set_backup_fs_root_level(root_backup,
1933                                btrfs_header_level(info->fs_root->node));
1934         }
1935
1936         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1937         btrfs_set_backup_dev_root_gen(root_backup,
1938                                btrfs_header_generation(info->dev_root->node));
1939         btrfs_set_backup_dev_root_level(root_backup,
1940                                        btrfs_header_level(info->dev_root->node));
1941
1942         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1943         btrfs_set_backup_csum_root_gen(root_backup,
1944                                btrfs_header_generation(info->csum_root->node));
1945         btrfs_set_backup_csum_root_level(root_backup,
1946                                btrfs_header_level(info->csum_root->node));
1947
1948         btrfs_set_backup_total_bytes(root_backup,
1949                              btrfs_super_total_bytes(info->super_copy));
1950         btrfs_set_backup_bytes_used(root_backup,
1951                              btrfs_super_bytes_used(info->super_copy));
1952         btrfs_set_backup_num_devices(root_backup,
1953                              btrfs_super_num_devices(info->super_copy));
1954
1955         /*
1956          * if we don't copy this out to the super_copy, it won't get remembered
1957          * for the next commit
1958          */
1959         memcpy(&info->super_copy->super_roots,
1960                &info->super_for_commit->super_roots,
1961                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1962 }
1963
1964 /*
1965  * this copies info out of the root backup array and back into
1966  * the in-memory super block.  It is meant to help iterate through
1967  * the array, so you send it the number of backups you've already
1968  * tried and the last backup index you used.
1969  *
1970  * this returns -1 when it has tried all the backups
1971  */
1972 static noinline int next_root_backup(struct btrfs_fs_info *info,
1973                                      struct btrfs_super_block *super,
1974                                      int *num_backups_tried, int *backup_index)
1975 {
1976         struct btrfs_root_backup *root_backup;
1977         int newest = *backup_index;
1978
1979         if (*num_backups_tried == 0) {
1980                 u64 gen = btrfs_super_generation(super);
1981
1982                 newest = find_newest_super_backup(info, gen);
1983                 if (newest == -1)
1984                         return -1;
1985
1986                 *backup_index = newest;
1987                 *num_backups_tried = 1;
1988         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1989                 /* we've tried all the backups, all done */
1990                 return -1;
1991         } else {
1992                 /* jump to the next oldest backup */
1993                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1994                         BTRFS_NUM_BACKUP_ROOTS;
1995                 *backup_index = newest;
1996                 *num_backups_tried += 1;
1997         }
1998         root_backup = super->super_roots + newest;
1999
2000         btrfs_set_super_generation(super,
2001                                    btrfs_backup_tree_root_gen(root_backup));
2002         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2003         btrfs_set_super_root_level(super,
2004                                    btrfs_backup_tree_root_level(root_backup));
2005         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2006
2007         /*
2008          * fixme: the total bytes and num_devices need to match or we should
2009          * need a fsck
2010          */
2011         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2012         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2013         return 0;
2014 }
2015
2016 /* helper to cleanup workers */
2017 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2018 {
2019         btrfs_destroy_workqueue(fs_info->fixup_workers);
2020         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2021         btrfs_destroy_workqueue(fs_info->workers);
2022         btrfs_destroy_workqueue(fs_info->endio_workers);
2023         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2024         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2025         btrfs_destroy_workqueue(fs_info->rmw_workers);
2026         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2027         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2028         btrfs_destroy_workqueue(fs_info->submit_workers);
2029         btrfs_destroy_workqueue(fs_info->delayed_workers);
2030         btrfs_destroy_workqueue(fs_info->caching_workers);
2031         btrfs_destroy_workqueue(fs_info->readahead_workers);
2032         btrfs_destroy_workqueue(fs_info->flush_workers);
2033         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2034         btrfs_destroy_workqueue(fs_info->extent_workers);
2035         /*
2036          * Now that all other work queues are destroyed, we can safely destroy
2037          * the queues used for metadata I/O, since tasks from those other work
2038          * queues can do metadata I/O operations.
2039          */
2040         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2041         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2042 }
2043
2044 static void free_root_extent_buffers(struct btrfs_root *root)
2045 {
2046         if (root) {
2047                 free_extent_buffer(root->node);
2048                 free_extent_buffer(root->commit_root);
2049                 root->node = NULL;
2050                 root->commit_root = NULL;
2051         }
2052 }
2053
2054 /* helper to cleanup tree roots */
2055 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2056 {
2057         free_root_extent_buffers(info->tree_root);
2058
2059         free_root_extent_buffers(info->dev_root);
2060         free_root_extent_buffers(info->extent_root);
2061         free_root_extent_buffers(info->csum_root);
2062         free_root_extent_buffers(info->quota_root);
2063         free_root_extent_buffers(info->uuid_root);
2064         if (chunk_root)
2065                 free_root_extent_buffers(info->chunk_root);
2066         free_root_extent_buffers(info->free_space_root);
2067 }
2068
2069 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2070 {
2071         int ret;
2072         struct btrfs_root *gang[8];
2073         int i;
2074
2075         while (!list_empty(&fs_info->dead_roots)) {
2076                 gang[0] = list_entry(fs_info->dead_roots.next,
2077                                      struct btrfs_root, root_list);
2078                 list_del(&gang[0]->root_list);
2079
2080                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2081                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2082                 } else {
2083                         free_extent_buffer(gang[0]->node);
2084                         free_extent_buffer(gang[0]->commit_root);
2085                         btrfs_put_fs_root(gang[0]);
2086                 }
2087         }
2088
2089         while (1) {
2090                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2091                                              (void **)gang, 0,
2092                                              ARRAY_SIZE(gang));
2093                 if (!ret)
2094                         break;
2095                 for (i = 0; i < ret; i++)
2096                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2097         }
2098
2099         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2100                 btrfs_free_log_root_tree(NULL, fs_info);
2101                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2102         }
2103 }
2104
2105 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2106 {
2107         mutex_init(&fs_info->scrub_lock);
2108         atomic_set(&fs_info->scrubs_running, 0);
2109         atomic_set(&fs_info->scrub_pause_req, 0);
2110         atomic_set(&fs_info->scrubs_paused, 0);
2111         atomic_set(&fs_info->scrub_cancel_req, 0);
2112         init_waitqueue_head(&fs_info->scrub_pause_wait);
2113         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2114 }
2115
2116 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2117 {
2118         spin_lock_init(&fs_info->balance_lock);
2119         mutex_init(&fs_info->balance_mutex);
2120         atomic_set(&fs_info->balance_pause_req, 0);
2121         atomic_set(&fs_info->balance_cancel_req, 0);
2122         fs_info->balance_ctl = NULL;
2123         init_waitqueue_head(&fs_info->balance_wait_q);
2124 }
2125
2126 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2127 {
2128         struct inode *inode = fs_info->btree_inode;
2129
2130         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2131         set_nlink(inode, 1);
2132         /*
2133          * we set the i_size on the btree inode to the max possible int.
2134          * the real end of the address space is determined by all of
2135          * the devices in the system
2136          */
2137         inode->i_size = OFFSET_MAX;
2138         inode->i_mapping->a_ops = &btree_aops;
2139
2140         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2141         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2142         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2143         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2144
2145         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2146
2147         BTRFS_I(inode)->root = fs_info->tree_root;
2148         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2149         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2150         btrfs_insert_inode_hash(inode);
2151 }
2152
2153 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2154 {
2155         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2156         init_rwsem(&fs_info->dev_replace.rwsem);
2157         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2158 }
2159
2160 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2161 {
2162         spin_lock_init(&fs_info->qgroup_lock);
2163         mutex_init(&fs_info->qgroup_ioctl_lock);
2164         fs_info->qgroup_tree = RB_ROOT;
2165         fs_info->qgroup_op_tree = RB_ROOT;
2166         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2167         fs_info->qgroup_seq = 1;
2168         fs_info->qgroup_ulist = NULL;
2169         fs_info->qgroup_rescan_running = false;
2170         mutex_init(&fs_info->qgroup_rescan_lock);
2171 }
2172
2173 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2174                 struct btrfs_fs_devices *fs_devices)
2175 {
2176         u32 max_active = fs_info->thread_pool_size;
2177         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2178
2179         fs_info->workers =
2180                 btrfs_alloc_workqueue(fs_info, "worker",
2181                                       flags | WQ_HIGHPRI, max_active, 16);
2182
2183         fs_info->delalloc_workers =
2184                 btrfs_alloc_workqueue(fs_info, "delalloc",
2185                                       flags, max_active, 2);
2186
2187         fs_info->flush_workers =
2188                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2189                                       flags, max_active, 0);
2190
2191         fs_info->caching_workers =
2192                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2193
2194         /*
2195          * a higher idle thresh on the submit workers makes it much more
2196          * likely that bios will be send down in a sane order to the
2197          * devices
2198          */
2199         fs_info->submit_workers =
2200                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2201                                       min_t(u64, fs_devices->num_devices,
2202                                             max_active), 64);
2203
2204         fs_info->fixup_workers =
2205                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2206
2207         /*
2208          * endios are largely parallel and should have a very
2209          * low idle thresh
2210          */
2211         fs_info->endio_workers =
2212                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2213         fs_info->endio_meta_workers =
2214                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2215                                       max_active, 4);
2216         fs_info->endio_meta_write_workers =
2217                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2218                                       max_active, 2);
2219         fs_info->endio_raid56_workers =
2220                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2221                                       max_active, 4);
2222         fs_info->endio_repair_workers =
2223                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2224         fs_info->rmw_workers =
2225                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2226         fs_info->endio_write_workers =
2227                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2228                                       max_active, 2);
2229         fs_info->endio_freespace_worker =
2230                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2231                                       max_active, 0);
2232         fs_info->delayed_workers =
2233                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2234                                       max_active, 0);
2235         fs_info->readahead_workers =
2236                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2237                                       max_active, 2);
2238         fs_info->qgroup_rescan_workers =
2239                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2240         fs_info->extent_workers =
2241                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2242                                       min_t(u64, fs_devices->num_devices,
2243                                             max_active), 8);
2244
2245         if (!(fs_info->workers && fs_info->delalloc_workers &&
2246               fs_info->submit_workers && fs_info->flush_workers &&
2247               fs_info->endio_workers && fs_info->endio_meta_workers &&
2248               fs_info->endio_meta_write_workers &&
2249               fs_info->endio_repair_workers &&
2250               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2251               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2252               fs_info->caching_workers && fs_info->readahead_workers &&
2253               fs_info->fixup_workers && fs_info->delayed_workers &&
2254               fs_info->extent_workers &&
2255               fs_info->qgroup_rescan_workers)) {
2256                 return -ENOMEM;
2257         }
2258
2259         return 0;
2260 }
2261
2262 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2263                             struct btrfs_fs_devices *fs_devices)
2264 {
2265         int ret;
2266         struct btrfs_root *log_tree_root;
2267         struct btrfs_super_block *disk_super = fs_info->super_copy;
2268         u64 bytenr = btrfs_super_log_root(disk_super);
2269         int level = btrfs_super_log_root_level(disk_super);
2270
2271         if (fs_devices->rw_devices == 0) {
2272                 btrfs_warn(fs_info, "log replay required on RO media");
2273                 return -EIO;
2274         }
2275
2276         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2277         if (!log_tree_root)
2278                 return -ENOMEM;
2279
2280         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2281
2282         log_tree_root->node = read_tree_block(fs_info, bytenr,
2283                                               fs_info->generation + 1,
2284                                               level, NULL);
2285         if (IS_ERR(log_tree_root->node)) {
2286                 btrfs_warn(fs_info, "failed to read log tree");
2287                 ret = PTR_ERR(log_tree_root->node);
2288                 kfree(log_tree_root);
2289                 return ret;
2290         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2291                 btrfs_err(fs_info, "failed to read log tree");
2292                 free_extent_buffer(log_tree_root->node);
2293                 kfree(log_tree_root);
2294                 return -EIO;
2295         }
2296         /* returns with log_tree_root freed on success */
2297         ret = btrfs_recover_log_trees(log_tree_root);
2298         if (ret) {
2299                 btrfs_handle_fs_error(fs_info, ret,
2300                                       "Failed to recover log tree");
2301                 free_extent_buffer(log_tree_root->node);
2302                 kfree(log_tree_root);
2303                 return ret;
2304         }
2305
2306         if (sb_rdonly(fs_info->sb)) {
2307                 ret = btrfs_commit_super(fs_info);
2308                 if (ret)
2309                         return ret;
2310         }
2311
2312         return 0;
2313 }
2314
2315 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2316 {
2317         struct btrfs_root *tree_root = fs_info->tree_root;
2318         struct btrfs_root *root;
2319         struct btrfs_key location;
2320         int ret;
2321
2322         BUG_ON(!fs_info->tree_root);
2323
2324         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2325         location.type = BTRFS_ROOT_ITEM_KEY;
2326         location.offset = 0;
2327
2328         root = btrfs_read_tree_root(tree_root, &location);
2329         if (IS_ERR(root)) {
2330                 ret = PTR_ERR(root);
2331                 goto out;
2332         }
2333         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2334         fs_info->extent_root = root;
2335
2336         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2337         root = btrfs_read_tree_root(tree_root, &location);
2338         if (IS_ERR(root)) {
2339                 ret = PTR_ERR(root);
2340                 goto out;
2341         }
2342         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2343         fs_info->dev_root = root;
2344         btrfs_init_devices_late(fs_info);
2345
2346         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2347         root = btrfs_read_tree_root(tree_root, &location);
2348         if (IS_ERR(root)) {
2349                 ret = PTR_ERR(root);
2350                 goto out;
2351         }
2352         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2353         fs_info->csum_root = root;
2354
2355         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2356         root = btrfs_read_tree_root(tree_root, &location);
2357         if (!IS_ERR(root)) {
2358                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2359                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2360                 fs_info->quota_root = root;
2361         }
2362
2363         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2364         root = btrfs_read_tree_root(tree_root, &location);
2365         if (IS_ERR(root)) {
2366                 ret = PTR_ERR(root);
2367                 if (ret != -ENOENT)
2368                         goto out;
2369         } else {
2370                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2371                 fs_info->uuid_root = root;
2372         }
2373
2374         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2375                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2376                 root = btrfs_read_tree_root(tree_root, &location);
2377                 if (IS_ERR(root)) {
2378                         ret = PTR_ERR(root);
2379                         goto out;
2380                 }
2381                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2382                 fs_info->free_space_root = root;
2383         }
2384
2385         return 0;
2386 out:
2387         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2388                    location.objectid, ret);
2389         return ret;
2390 }
2391
2392 /*
2393  * Real super block validation
2394  * NOTE: super csum type and incompat features will not be checked here.
2395  *
2396  * @sb:         super block to check
2397  * @mirror_num: the super block number to check its bytenr:
2398  *              0       the primary (1st) sb
2399  *              1, 2    2nd and 3rd backup copy
2400  *             -1       skip bytenr check
2401  */
2402 static int validate_super(struct btrfs_fs_info *fs_info,
2403                             struct btrfs_super_block *sb, int mirror_num)
2404 {
2405         u64 nodesize = btrfs_super_nodesize(sb);
2406         u64 sectorsize = btrfs_super_sectorsize(sb);
2407         int ret = 0;
2408
2409         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2410                 btrfs_err(fs_info, "no valid FS found");
2411                 ret = -EINVAL;
2412         }
2413         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2414                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2415                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2416                 ret = -EINVAL;
2417         }
2418         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2419                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2420                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2421                 ret = -EINVAL;
2422         }
2423         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2424                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2425                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2426                 ret = -EINVAL;
2427         }
2428         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2429                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2430                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2431                 ret = -EINVAL;
2432         }
2433
2434         /*
2435          * Check sectorsize and nodesize first, other check will need it.
2436          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2437          */
2438         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2439             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2440                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2441                 ret = -EINVAL;
2442         }
2443         /* Only PAGE SIZE is supported yet */
2444         if (sectorsize != PAGE_SIZE) {
2445                 btrfs_err(fs_info,
2446                         "sectorsize %llu not supported yet, only support %lu",
2447                         sectorsize, PAGE_SIZE);
2448                 ret = -EINVAL;
2449         }
2450         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2451             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2452                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2453                 ret = -EINVAL;
2454         }
2455         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2456                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2457                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2458                 ret = -EINVAL;
2459         }
2460
2461         /* Root alignment check */
2462         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2463                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2464                            btrfs_super_root(sb));
2465                 ret = -EINVAL;
2466         }
2467         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2468                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2469                            btrfs_super_chunk_root(sb));
2470                 ret = -EINVAL;
2471         }
2472         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2473                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2474                            btrfs_super_log_root(sb));
2475                 ret = -EINVAL;
2476         }
2477
2478         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2479                    BTRFS_FSID_SIZE) != 0) {
2480                 btrfs_err(fs_info,
2481                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2482                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2483                 ret = -EINVAL;
2484         }
2485
2486         /*
2487          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2488          * done later
2489          */
2490         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2491                 btrfs_err(fs_info, "bytes_used is too small %llu",
2492                           btrfs_super_bytes_used(sb));
2493                 ret = -EINVAL;
2494         }
2495         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2496                 btrfs_err(fs_info, "invalid stripesize %u",
2497                           btrfs_super_stripesize(sb));
2498                 ret = -EINVAL;
2499         }
2500         if (btrfs_super_num_devices(sb) > (1UL << 31))
2501                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2502                            btrfs_super_num_devices(sb));
2503         if (btrfs_super_num_devices(sb) == 0) {
2504                 btrfs_err(fs_info, "number of devices is 0");
2505                 ret = -EINVAL;
2506         }
2507
2508         if (mirror_num >= 0 &&
2509             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2510                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2511                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2512                 ret = -EINVAL;
2513         }
2514
2515         /*
2516          * Obvious sys_chunk_array corruptions, it must hold at least one key
2517          * and one chunk
2518          */
2519         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2520                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2521                           btrfs_super_sys_array_size(sb),
2522                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2523                 ret = -EINVAL;
2524         }
2525         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2526                         + sizeof(struct btrfs_chunk)) {
2527                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2528                           btrfs_super_sys_array_size(sb),
2529                           sizeof(struct btrfs_disk_key)
2530                           + sizeof(struct btrfs_chunk));
2531                 ret = -EINVAL;
2532         }
2533
2534         /*
2535          * The generation is a global counter, we'll trust it more than the others
2536          * but it's still possible that it's the one that's wrong.
2537          */
2538         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2539                 btrfs_warn(fs_info,
2540                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2541                         btrfs_super_generation(sb),
2542                         btrfs_super_chunk_root_generation(sb));
2543         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2544             && btrfs_super_cache_generation(sb) != (u64)-1)
2545                 btrfs_warn(fs_info,
2546                         "suspicious: generation < cache_generation: %llu < %llu",
2547                         btrfs_super_generation(sb),
2548                         btrfs_super_cache_generation(sb));
2549
2550         return ret;
2551 }
2552
2553 /*
2554  * Validation of super block at mount time.
2555  * Some checks already done early at mount time, like csum type and incompat
2556  * flags will be skipped.
2557  */
2558 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2559 {
2560         return validate_super(fs_info, fs_info->super_copy, 0);
2561 }
2562
2563 /*
2564  * Validation of super block at write time.
2565  * Some checks like bytenr check will be skipped as their values will be
2566  * overwritten soon.
2567  * Extra checks like csum type and incompat flags will be done here.
2568  */
2569 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2570                                       struct btrfs_super_block *sb)
2571 {
2572         int ret;
2573
2574         ret = validate_super(fs_info, sb, -1);
2575         if (ret < 0)
2576                 goto out;
2577         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2578                 ret = -EUCLEAN;
2579                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2580                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2581                 goto out;
2582         }
2583         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2584                 ret = -EUCLEAN;
2585                 btrfs_err(fs_info,
2586                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2587                           btrfs_super_incompat_flags(sb),
2588                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2589                 goto out;
2590         }
2591 out:
2592         if (ret < 0)
2593                 btrfs_err(fs_info,
2594                 "super block corruption detected before writing it to disk");
2595         return ret;
2596 }
2597
2598 int open_ctree(struct super_block *sb,
2599                struct btrfs_fs_devices *fs_devices,
2600                char *options)
2601 {
2602         u32 sectorsize;
2603         u32 nodesize;
2604         u32 stripesize;
2605         u64 generation;
2606         u64 features;
2607         struct btrfs_key location;
2608         struct buffer_head *bh;
2609         struct btrfs_super_block *disk_super;
2610         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2611         struct btrfs_root *tree_root;
2612         struct btrfs_root *chunk_root;
2613         int ret;
2614         int err = -EINVAL;
2615         int num_backups_tried = 0;
2616         int backup_index = 0;
2617         int clear_free_space_tree = 0;
2618         int level;
2619
2620         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2621         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2622         if (!tree_root || !chunk_root) {
2623                 err = -ENOMEM;
2624                 goto fail;
2625         }
2626
2627         ret = init_srcu_struct(&fs_info->subvol_srcu);
2628         if (ret) {
2629                 err = ret;
2630                 goto fail;
2631         }
2632
2633         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2634         if (ret) {
2635                 err = ret;
2636                 goto fail_srcu;
2637         }
2638         fs_info->dirty_metadata_batch = PAGE_SIZE *
2639                                         (1 + ilog2(nr_cpu_ids));
2640
2641         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2642         if (ret) {
2643                 err = ret;
2644                 goto fail_dirty_metadata_bytes;
2645         }
2646
2647         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2648                         GFP_KERNEL);
2649         if (ret) {
2650                 err = ret;
2651                 goto fail_delalloc_bytes;
2652         }
2653
2654         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2655         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2656         INIT_LIST_HEAD(&fs_info->trans_list);
2657         INIT_LIST_HEAD(&fs_info->dead_roots);
2658         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2659         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2660         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2661         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2662         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2663         spin_lock_init(&fs_info->delalloc_root_lock);
2664         spin_lock_init(&fs_info->trans_lock);
2665         spin_lock_init(&fs_info->fs_roots_radix_lock);
2666         spin_lock_init(&fs_info->delayed_iput_lock);
2667         spin_lock_init(&fs_info->defrag_inodes_lock);
2668         spin_lock_init(&fs_info->tree_mod_seq_lock);
2669         spin_lock_init(&fs_info->super_lock);
2670         spin_lock_init(&fs_info->qgroup_op_lock);
2671         spin_lock_init(&fs_info->buffer_lock);
2672         spin_lock_init(&fs_info->unused_bgs_lock);
2673         rwlock_init(&fs_info->tree_mod_log_lock);
2674         mutex_init(&fs_info->unused_bg_unpin_mutex);
2675         mutex_init(&fs_info->delete_unused_bgs_mutex);
2676         mutex_init(&fs_info->reloc_mutex);
2677         mutex_init(&fs_info->delalloc_root_mutex);
2678         seqlock_init(&fs_info->profiles_lock);
2679
2680         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2681         INIT_LIST_HEAD(&fs_info->space_info);
2682         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2683         INIT_LIST_HEAD(&fs_info->unused_bgs);
2684         btrfs_mapping_init(&fs_info->mapping_tree);
2685         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2686                              BTRFS_BLOCK_RSV_GLOBAL);
2687         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2688         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2689         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2690         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2691                              BTRFS_BLOCK_RSV_DELOPS);
2692         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2693                              BTRFS_BLOCK_RSV_DELREFS);
2694
2695         atomic_set(&fs_info->async_delalloc_pages, 0);
2696         atomic_set(&fs_info->defrag_running, 0);
2697         atomic_set(&fs_info->qgroup_op_seq, 0);
2698         atomic_set(&fs_info->reada_works_cnt, 0);
2699         atomic_set(&fs_info->nr_delayed_iputs, 0);
2700         atomic64_set(&fs_info->tree_mod_seq, 0);
2701         fs_info->sb = sb;
2702         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2703         fs_info->metadata_ratio = 0;
2704         fs_info->defrag_inodes = RB_ROOT;
2705         atomic64_set(&fs_info->free_chunk_space, 0);
2706         fs_info->tree_mod_log = RB_ROOT;
2707         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2708         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2709         /* readahead state */
2710         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2711         spin_lock_init(&fs_info->reada_lock);
2712         btrfs_init_ref_verify(fs_info);
2713
2714         fs_info->thread_pool_size = min_t(unsigned long,
2715                                           num_online_cpus() + 2, 8);
2716
2717         INIT_LIST_HEAD(&fs_info->ordered_roots);
2718         spin_lock_init(&fs_info->ordered_root_lock);
2719
2720         fs_info->btree_inode = new_inode(sb);
2721         if (!fs_info->btree_inode) {
2722                 err = -ENOMEM;
2723                 goto fail_bio_counter;
2724         }
2725         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2726
2727         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2728                                         GFP_KERNEL);
2729         if (!fs_info->delayed_root) {
2730                 err = -ENOMEM;
2731                 goto fail_iput;
2732         }
2733         btrfs_init_delayed_root(fs_info->delayed_root);
2734
2735         btrfs_init_scrub(fs_info);
2736 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2737         fs_info->check_integrity_print_mask = 0;
2738 #endif
2739         btrfs_init_balance(fs_info);
2740         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2741
2742         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2743         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2744
2745         btrfs_init_btree_inode(fs_info);
2746
2747         spin_lock_init(&fs_info->block_group_cache_lock);
2748         fs_info->block_group_cache_tree = RB_ROOT;
2749         fs_info->first_logical_byte = (u64)-1;
2750
2751         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2752         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2753         fs_info->pinned_extents = &fs_info->freed_extents[0];
2754         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2755
2756         mutex_init(&fs_info->ordered_operations_mutex);
2757         mutex_init(&fs_info->tree_log_mutex);
2758         mutex_init(&fs_info->chunk_mutex);
2759         mutex_init(&fs_info->transaction_kthread_mutex);
2760         mutex_init(&fs_info->cleaner_mutex);
2761         mutex_init(&fs_info->ro_block_group_mutex);
2762         init_rwsem(&fs_info->commit_root_sem);
2763         init_rwsem(&fs_info->cleanup_work_sem);
2764         init_rwsem(&fs_info->subvol_sem);
2765         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2766
2767         btrfs_init_dev_replace_locks(fs_info);
2768         btrfs_init_qgroup(fs_info);
2769
2770         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2771         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2772
2773         init_waitqueue_head(&fs_info->transaction_throttle);
2774         init_waitqueue_head(&fs_info->transaction_wait);
2775         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2776         init_waitqueue_head(&fs_info->async_submit_wait);
2777         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2778
2779         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2780
2781         /* Usable values until the real ones are cached from the superblock */
2782         fs_info->nodesize = 4096;
2783         fs_info->sectorsize = 4096;
2784         fs_info->stripesize = 4096;
2785
2786         spin_lock_init(&fs_info->swapfile_pins_lock);
2787         fs_info->swapfile_pins = RB_ROOT;
2788
2789         ret = btrfs_alloc_stripe_hash_table(fs_info);
2790         if (ret) {
2791                 err = ret;
2792                 goto fail_alloc;
2793         }
2794
2795         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2796
2797         invalidate_bdev(fs_devices->latest_bdev);
2798
2799         /*
2800          * Read super block and check the signature bytes only
2801          */
2802         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2803         if (IS_ERR(bh)) {
2804                 err = PTR_ERR(bh);
2805                 goto fail_alloc;
2806         }
2807
2808         /*
2809          * We want to check superblock checksum, the type is stored inside.
2810          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2811          */
2812         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2813                 btrfs_err(fs_info, "superblock checksum mismatch");
2814                 err = -EINVAL;
2815                 brelse(bh);
2816                 goto fail_alloc;
2817         }
2818
2819         /*
2820          * super_copy is zeroed at allocation time and we never touch the
2821          * following bytes up to INFO_SIZE, the checksum is calculated from
2822          * the whole block of INFO_SIZE
2823          */
2824         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2825         brelse(bh);
2826
2827         disk_super = fs_info->super_copy;
2828
2829         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2830                        BTRFS_FSID_SIZE));
2831
2832         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2833                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2834                                 fs_info->super_copy->metadata_uuid,
2835                                 BTRFS_FSID_SIZE));
2836         }
2837
2838         features = btrfs_super_flags(disk_super);
2839         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2840                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2841                 btrfs_set_super_flags(disk_super, features);
2842                 btrfs_info(fs_info,
2843                         "found metadata UUID change in progress flag, clearing");
2844         }
2845
2846         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2847                sizeof(*fs_info->super_for_commit));
2848
2849         ret = btrfs_validate_mount_super(fs_info);
2850         if (ret) {
2851                 btrfs_err(fs_info, "superblock contains fatal errors");
2852                 err = -EINVAL;
2853                 goto fail_alloc;
2854         }
2855
2856         if (!btrfs_super_root(disk_super))
2857                 goto fail_alloc;
2858
2859         /* check FS state, whether FS is broken. */
2860         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2861                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2862
2863         /*
2864          * run through our array of backup supers and setup
2865          * our ring pointer to the oldest one
2866          */
2867         generation = btrfs_super_generation(disk_super);
2868         find_oldest_super_backup(fs_info, generation);
2869
2870         /*
2871          * In the long term, we'll store the compression type in the super
2872          * block, and it'll be used for per file compression control.
2873          */
2874         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2875
2876         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2877         if (ret) {
2878                 err = ret;
2879                 goto fail_alloc;
2880         }
2881
2882         features = btrfs_super_incompat_flags(disk_super) &
2883                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2884         if (features) {
2885                 btrfs_err(fs_info,
2886                     "cannot mount because of unsupported optional features (%llx)",
2887                     features);
2888                 err = -EINVAL;
2889                 goto fail_alloc;
2890         }
2891
2892         features = btrfs_super_incompat_flags(disk_super);
2893         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2894         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2895                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2896         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2897                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2898
2899         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2900                 btrfs_info(fs_info, "has skinny extents");
2901
2902         /*
2903          * flag our filesystem as having big metadata blocks if
2904          * they are bigger than the page size
2905          */
2906         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2907                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2908                         btrfs_info(fs_info,
2909                                 "flagging fs with big metadata feature");
2910                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2911         }
2912
2913         nodesize = btrfs_super_nodesize(disk_super);
2914         sectorsize = btrfs_super_sectorsize(disk_super);
2915         stripesize = sectorsize;
2916         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2917         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2918
2919         /* Cache block sizes */
2920         fs_info->nodesize = nodesize;
2921         fs_info->sectorsize = sectorsize;
2922         fs_info->stripesize = stripesize;
2923
2924         /*
2925          * mixed block groups end up with duplicate but slightly offset
2926          * extent buffers for the same range.  It leads to corruptions
2927          */
2928         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2929             (sectorsize != nodesize)) {
2930                 btrfs_err(fs_info,
2931 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2932                         nodesize, sectorsize);
2933                 goto fail_alloc;
2934         }
2935
2936         /*
2937          * Needn't use the lock because there is no other task which will
2938          * update the flag.
2939          */
2940         btrfs_set_super_incompat_flags(disk_super, features);
2941
2942         features = btrfs_super_compat_ro_flags(disk_super) &
2943                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2944         if (!sb_rdonly(sb) && features) {
2945                 btrfs_err(fs_info,
2946         "cannot mount read-write because of unsupported optional features (%llx)",
2947                        features);
2948                 err = -EINVAL;
2949                 goto fail_alloc;
2950         }
2951
2952         ret = btrfs_init_workqueues(fs_info, fs_devices);
2953         if (ret) {
2954                 err = ret;
2955                 goto fail_sb_buffer;
2956         }
2957
2958         sb->s_bdi->congested_fn = btrfs_congested_fn;
2959         sb->s_bdi->congested_data = fs_info;
2960         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2961         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2962         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2963         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2964
2965         sb->s_blocksize = sectorsize;
2966         sb->s_blocksize_bits = blksize_bits(sectorsize);
2967         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2968
2969         mutex_lock(&fs_info->chunk_mutex);
2970         ret = btrfs_read_sys_array(fs_info);
2971         mutex_unlock(&fs_info->chunk_mutex);
2972         if (ret) {
2973                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2974                 goto fail_sb_buffer;
2975         }
2976
2977         generation = btrfs_super_chunk_root_generation(disk_super);
2978         level = btrfs_super_chunk_root_level(disk_super);
2979
2980         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2981
2982         chunk_root->node = read_tree_block(fs_info,
2983                                            btrfs_super_chunk_root(disk_super),
2984                                            generation, level, NULL);
2985         if (IS_ERR(chunk_root->node) ||
2986             !extent_buffer_uptodate(chunk_root->node)) {
2987                 btrfs_err(fs_info, "failed to read chunk root");
2988                 if (!IS_ERR(chunk_root->node))
2989                         free_extent_buffer(chunk_root->node);
2990                 chunk_root->node = NULL;
2991                 goto fail_tree_roots;
2992         }
2993         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2994         chunk_root->commit_root = btrfs_root_node(chunk_root);
2995
2996         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2997            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2998
2999         ret = btrfs_read_chunk_tree(fs_info);
3000         if (ret) {
3001                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3002                 goto fail_tree_roots;
3003         }
3004
3005         /*
3006          * Keep the devid that is marked to be the target device for the
3007          * device replace procedure
3008          */
3009         btrfs_free_extra_devids(fs_devices, 0);
3010
3011         if (!fs_devices->latest_bdev) {
3012                 btrfs_err(fs_info, "failed to read devices");
3013                 goto fail_tree_roots;
3014         }
3015
3016 retry_root_backup:
3017         generation = btrfs_super_generation(disk_super);
3018         level = btrfs_super_root_level(disk_super);
3019
3020         tree_root->node = read_tree_block(fs_info,
3021                                           btrfs_super_root(disk_super),
3022                                           generation, level, NULL);
3023         if (IS_ERR(tree_root->node) ||
3024             !extent_buffer_uptodate(tree_root->node)) {
3025                 btrfs_warn(fs_info, "failed to read tree root");
3026                 if (!IS_ERR(tree_root->node))
3027                         free_extent_buffer(tree_root->node);
3028                 tree_root->node = NULL;
3029                 goto recovery_tree_root;
3030         }
3031
3032         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3033         tree_root->commit_root = btrfs_root_node(tree_root);
3034         btrfs_set_root_refs(&tree_root->root_item, 1);
3035
3036         mutex_lock(&tree_root->objectid_mutex);
3037         ret = btrfs_find_highest_objectid(tree_root,
3038                                         &tree_root->highest_objectid);
3039         if (ret) {
3040                 mutex_unlock(&tree_root->objectid_mutex);
3041                 goto recovery_tree_root;
3042         }
3043
3044         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3045
3046         mutex_unlock(&tree_root->objectid_mutex);
3047
3048         ret = btrfs_read_roots(fs_info);
3049         if (ret)
3050                 goto recovery_tree_root;
3051
3052         fs_info->generation = generation;
3053         fs_info->last_trans_committed = generation;
3054
3055         ret = btrfs_verify_dev_extents(fs_info);
3056         if (ret) {
3057                 btrfs_err(fs_info,
3058                           "failed to verify dev extents against chunks: %d",
3059                           ret);
3060                 goto fail_block_groups;
3061         }
3062         ret = btrfs_recover_balance(fs_info);
3063         if (ret) {
3064                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3065                 goto fail_block_groups;
3066         }
3067
3068         ret = btrfs_init_dev_stats(fs_info);
3069         if (ret) {
3070                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3071                 goto fail_block_groups;
3072         }
3073
3074         ret = btrfs_init_dev_replace(fs_info);
3075         if (ret) {
3076                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3077                 goto fail_block_groups;
3078         }
3079
3080         btrfs_free_extra_devids(fs_devices, 1);
3081
3082         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3083         if (ret) {
3084                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3085                                 ret);
3086                 goto fail_block_groups;
3087         }
3088
3089         ret = btrfs_sysfs_add_device(fs_devices);
3090         if (ret) {
3091                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3092                                 ret);
3093                 goto fail_fsdev_sysfs;
3094         }
3095
3096         ret = btrfs_sysfs_add_mounted(fs_info);
3097         if (ret) {
3098                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3099                 goto fail_fsdev_sysfs;
3100         }
3101
3102         ret = btrfs_init_space_info(fs_info);
3103         if (ret) {
3104                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3105                 goto fail_sysfs;
3106         }
3107
3108         ret = btrfs_read_block_groups(fs_info);
3109         if (ret) {
3110                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3111                 goto fail_sysfs;
3112         }
3113
3114         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3115                 btrfs_warn(fs_info,
3116                 "writable mount is not allowed due to too many missing devices");
3117                 goto fail_sysfs;
3118         }
3119
3120         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3121                                                "btrfs-cleaner");
3122         if (IS_ERR(fs_info->cleaner_kthread))
3123                 goto fail_sysfs;
3124
3125         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3126                                                    tree_root,
3127                                                    "btrfs-transaction");
3128         if (IS_ERR(fs_info->transaction_kthread))
3129                 goto fail_cleaner;
3130
3131         if (!btrfs_test_opt(fs_info, NOSSD) &&
3132             !fs_info->fs_devices->rotating) {
3133                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3134         }
3135
3136         /*
3137          * Mount does not set all options immediately, we can do it now and do
3138          * not have to wait for transaction commit
3139          */
3140         btrfs_apply_pending_changes(fs_info);
3141
3142 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3143         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3144                 ret = btrfsic_mount(fs_info, fs_devices,
3145                                     btrfs_test_opt(fs_info,
3146                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3147                                     1 : 0,
3148                                     fs_info->check_integrity_print_mask);
3149                 if (ret)
3150                         btrfs_warn(fs_info,
3151                                 "failed to initialize integrity check module: %d",
3152                                 ret);
3153         }
3154 #endif
3155         ret = btrfs_read_qgroup_config(fs_info);
3156         if (ret)
3157                 goto fail_trans_kthread;
3158
3159         if (btrfs_build_ref_tree(fs_info))
3160                 btrfs_err(fs_info, "couldn't build ref tree");
3161
3162         /* do not make disk changes in broken FS or nologreplay is given */
3163         if (btrfs_super_log_root(disk_super) != 0 &&
3164             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3165                 ret = btrfs_replay_log(fs_info, fs_devices);
3166                 if (ret) {
3167                         err = ret;
3168                         goto fail_qgroup;
3169                 }
3170         }
3171
3172         ret = btrfs_find_orphan_roots(fs_info);
3173         if (ret)
3174                 goto fail_qgroup;
3175
3176         if (!sb_rdonly(sb)) {
3177                 ret = btrfs_cleanup_fs_roots(fs_info);
3178                 if (ret)
3179                         goto fail_qgroup;
3180
3181                 mutex_lock(&fs_info->cleaner_mutex);
3182                 ret = btrfs_recover_relocation(tree_root);
3183                 mutex_unlock(&fs_info->cleaner_mutex);
3184                 if (ret < 0) {
3185                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3186                                         ret);
3187                         err = -EINVAL;
3188                         goto fail_qgroup;
3189                 }
3190         }
3191
3192         location.objectid = BTRFS_FS_TREE_OBJECTID;
3193         location.type = BTRFS_ROOT_ITEM_KEY;
3194         location.offset = 0;
3195
3196         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3197         if (IS_ERR(fs_info->fs_root)) {
3198                 err = PTR_ERR(fs_info->fs_root);
3199                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3200                 goto fail_qgroup;
3201         }
3202
3203         if (sb_rdonly(sb))
3204                 return 0;
3205
3206         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3207             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3208                 clear_free_space_tree = 1;
3209         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3210                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3211                 btrfs_warn(fs_info, "free space tree is invalid");
3212                 clear_free_space_tree = 1;
3213         }
3214
3215         if (clear_free_space_tree) {
3216                 btrfs_info(fs_info, "clearing free space tree");
3217                 ret = btrfs_clear_free_space_tree(fs_info);
3218                 if (ret) {
3219                         btrfs_warn(fs_info,
3220                                    "failed to clear free space tree: %d", ret);
3221                         close_ctree(fs_info);
3222                         return ret;
3223                 }
3224         }
3225
3226         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3227             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3228                 btrfs_info(fs_info, "creating free space tree");
3229                 ret = btrfs_create_free_space_tree(fs_info);
3230                 if (ret) {
3231                         btrfs_warn(fs_info,
3232                                 "failed to create free space tree: %d", ret);
3233                         close_ctree(fs_info);
3234                         return ret;
3235                 }
3236         }
3237
3238         down_read(&fs_info->cleanup_work_sem);
3239         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3240             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3241                 up_read(&fs_info->cleanup_work_sem);
3242                 close_ctree(fs_info);
3243                 return ret;
3244         }
3245         up_read(&fs_info->cleanup_work_sem);
3246
3247         ret = btrfs_resume_balance_async(fs_info);
3248         if (ret) {
3249                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3250                 close_ctree(fs_info);
3251                 return ret;
3252         }
3253
3254         ret = btrfs_resume_dev_replace_async(fs_info);
3255         if (ret) {
3256                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3257                 close_ctree(fs_info);
3258                 return ret;
3259         }
3260
3261         btrfs_qgroup_rescan_resume(fs_info);
3262
3263         if (!fs_info->uuid_root) {
3264                 btrfs_info(fs_info, "creating UUID tree");
3265                 ret = btrfs_create_uuid_tree(fs_info);
3266                 if (ret) {
3267                         btrfs_warn(fs_info,
3268                                 "failed to create the UUID tree: %d", ret);
3269                         close_ctree(fs_info);
3270                         return ret;
3271                 }
3272         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3273                    fs_info->generation !=
3274                                 btrfs_super_uuid_tree_generation(disk_super)) {
3275                 btrfs_info(fs_info, "checking UUID tree");
3276                 ret = btrfs_check_uuid_tree(fs_info);
3277                 if (ret) {
3278                         btrfs_warn(fs_info,
3279                                 "failed to check the UUID tree: %d", ret);
3280                         close_ctree(fs_info);
3281                         return ret;
3282                 }
3283         } else {
3284                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3285         }
3286         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3287
3288         /*
3289          * backuproot only affect mount behavior, and if open_ctree succeeded,
3290          * no need to keep the flag
3291          */
3292         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3293
3294         return 0;
3295
3296 fail_qgroup:
3297         btrfs_free_qgroup_config(fs_info);
3298 fail_trans_kthread:
3299         kthread_stop(fs_info->transaction_kthread);
3300         btrfs_cleanup_transaction(fs_info);
3301         btrfs_free_fs_roots(fs_info);
3302 fail_cleaner:
3303         kthread_stop(fs_info->cleaner_kthread);
3304
3305         /*
3306          * make sure we're done with the btree inode before we stop our
3307          * kthreads
3308          */
3309         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3310
3311 fail_sysfs:
3312         btrfs_sysfs_remove_mounted(fs_info);
3313
3314 fail_fsdev_sysfs:
3315         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3316
3317 fail_block_groups:
3318         btrfs_put_block_group_cache(fs_info);
3319
3320 fail_tree_roots:
3321         free_root_pointers(fs_info, 1);
3322         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3323
3324 fail_sb_buffer:
3325         btrfs_stop_all_workers(fs_info);
3326         btrfs_free_block_groups(fs_info);
3327 fail_alloc:
3328 fail_iput:
3329         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3330
3331         iput(fs_info->btree_inode);
3332 fail_bio_counter:
3333         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
3334 fail_delalloc_bytes:
3335         percpu_counter_destroy(&fs_info->delalloc_bytes);
3336 fail_dirty_metadata_bytes:
3337         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3338 fail_srcu:
3339         cleanup_srcu_struct(&fs_info->subvol_srcu);
3340 fail:
3341         btrfs_free_stripe_hash_table(fs_info);
3342         btrfs_close_devices(fs_info->fs_devices);
3343         return err;
3344
3345 recovery_tree_root:
3346         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3347                 goto fail_tree_roots;
3348
3349         free_root_pointers(fs_info, 0);
3350
3351         /* don't use the log in recovery mode, it won't be valid */
3352         btrfs_set_super_log_root(disk_super, 0);
3353
3354         /* we can't trust the free space cache either */
3355         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3356
3357         ret = next_root_backup(fs_info, fs_info->super_copy,
3358                                &num_backups_tried, &backup_index);
3359         if (ret == -1)
3360                 goto fail_block_groups;
3361         goto retry_root_backup;
3362 }
3363 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3364
3365 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3366 {
3367         if (uptodate) {
3368                 set_buffer_uptodate(bh);
3369         } else {
3370                 struct btrfs_device *device = (struct btrfs_device *)
3371                         bh->b_private;
3372
3373                 btrfs_warn_rl_in_rcu(device->fs_info,
3374                                 "lost page write due to IO error on %s",
3375                                           rcu_str_deref(device->name));
3376                 /* note, we don't set_buffer_write_io_error because we have
3377                  * our own ways of dealing with the IO errors
3378                  */
3379                 clear_buffer_uptodate(bh);
3380                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3381         }
3382         unlock_buffer(bh);
3383         put_bh(bh);
3384 }
3385
3386 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3387                         struct buffer_head **bh_ret)
3388 {
3389         struct buffer_head *bh;
3390         struct btrfs_super_block *super;
3391         u64 bytenr;
3392
3393         bytenr = btrfs_sb_offset(copy_num);
3394         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3395                 return -EINVAL;
3396
3397         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3398         /*
3399          * If we fail to read from the underlying devices, as of now
3400          * the best option we have is to mark it EIO.
3401          */
3402         if (!bh)
3403                 return -EIO;
3404
3405         super = (struct btrfs_super_block *)bh->b_data;
3406         if (btrfs_super_bytenr(super) != bytenr ||
3407                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3408                 brelse(bh);
3409                 return -EINVAL;
3410         }
3411
3412         *bh_ret = bh;
3413         return 0;
3414 }
3415
3416
3417 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3418 {
3419         struct buffer_head *bh;
3420         struct buffer_head *latest = NULL;
3421         struct btrfs_super_block *super;
3422         int i;
3423         u64 transid = 0;
3424         int ret = -EINVAL;
3425
3426         /* we would like to check all the supers, but that would make
3427          * a btrfs mount succeed after a mkfs from a different FS.
3428          * So, we need to add a special mount option to scan for
3429          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3430          */
3431         for (i = 0; i < 1; i++) {
3432                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3433                 if (ret)
3434                         continue;
3435
3436                 super = (struct btrfs_super_block *)bh->b_data;
3437
3438                 if (!latest || btrfs_super_generation(super) > transid) {
3439                         brelse(latest);
3440                         latest = bh;
3441                         transid = btrfs_super_generation(super);
3442                 } else {
3443                         brelse(bh);
3444                 }
3445         }
3446
3447         if (!latest)
3448                 return ERR_PTR(ret);
3449
3450         return latest;
3451 }
3452
3453 /*
3454  * Write superblock @sb to the @device. Do not wait for completion, all the
3455  * buffer heads we write are pinned.
3456  *
3457  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3458  * the expected device size at commit time. Note that max_mirrors must be
3459  * same for write and wait phases.
3460  *
3461  * Return number of errors when buffer head is not found or submission fails.
3462  */
3463 static int write_dev_supers(struct btrfs_device *device,
3464                             struct btrfs_super_block *sb, int max_mirrors)
3465 {
3466         struct buffer_head *bh;
3467         int i;
3468         int ret;
3469         int errors = 0;
3470         u32 crc;
3471         u64 bytenr;
3472         int op_flags;
3473
3474         if (max_mirrors == 0)
3475                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3476
3477         for (i = 0; i < max_mirrors; i++) {
3478                 bytenr = btrfs_sb_offset(i);
3479                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3480                     device->commit_total_bytes)
3481                         break;
3482
3483                 btrfs_set_super_bytenr(sb, bytenr);
3484
3485                 crc = ~(u32)0;
3486                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3487                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3488                 btrfs_csum_final(crc, sb->csum);
3489
3490                 /* One reference for us, and we leave it for the caller */
3491                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3492                               BTRFS_SUPER_INFO_SIZE);
3493                 if (!bh) {
3494                         btrfs_err(device->fs_info,
3495                             "couldn't get super buffer head for bytenr %llu",
3496                             bytenr);
3497                         errors++;
3498                         continue;
3499                 }
3500
3501                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3502
3503                 /* one reference for submit_bh */
3504                 get_bh(bh);
3505
3506                 set_buffer_uptodate(bh);
3507                 lock_buffer(bh);
3508                 bh->b_end_io = btrfs_end_buffer_write_sync;
3509                 bh->b_private = device;
3510
3511                 /*
3512                  * we fua the first super.  The others we allow
3513                  * to go down lazy.
3514                  */
3515                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3516                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3517                         op_flags |= REQ_FUA;
3518                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3519                 if (ret)
3520                         errors++;
3521         }
3522         return errors < i ? 0 : -1;
3523 }
3524
3525 /*
3526  * Wait for write completion of superblocks done by write_dev_supers,
3527  * @max_mirrors same for write and wait phases.
3528  *
3529  * Return number of errors when buffer head is not found or not marked up to
3530  * date.
3531  */
3532 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3533 {
3534         struct buffer_head *bh;
3535         int i;
3536         int errors = 0;
3537         bool primary_failed = false;
3538         u64 bytenr;
3539
3540         if (max_mirrors == 0)
3541                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3542
3543         for (i = 0; i < max_mirrors; i++) {
3544                 bytenr = btrfs_sb_offset(i);
3545                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3546                     device->commit_total_bytes)
3547                         break;
3548
3549                 bh = __find_get_block(device->bdev,
3550                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3551                                       BTRFS_SUPER_INFO_SIZE);
3552                 if (!bh) {
3553                         errors++;
3554                         if (i == 0)
3555                                 primary_failed = true;
3556                         continue;
3557                 }
3558                 wait_on_buffer(bh);
3559                 if (!buffer_uptodate(bh)) {
3560                         errors++;
3561                         if (i == 0)
3562                                 primary_failed = true;
3563                 }
3564
3565                 /* drop our reference */
3566                 brelse(bh);
3567
3568                 /* drop the reference from the writing run */
3569                 brelse(bh);
3570         }
3571
3572         /* log error, force error return */
3573         if (primary_failed) {
3574                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3575                           device->devid);
3576                 return -1;
3577         }
3578
3579         return errors < i ? 0 : -1;
3580 }
3581
3582 /*
3583  * endio for the write_dev_flush, this will wake anyone waiting
3584  * for the barrier when it is done
3585  */
3586 static void btrfs_end_empty_barrier(struct bio *bio)
3587 {
3588         complete(bio->bi_private);
3589 }
3590
3591 /*
3592  * Submit a flush request to the device if it supports it. Error handling is
3593  * done in the waiting counterpart.
3594  */
3595 static void write_dev_flush(struct btrfs_device *device)
3596 {
3597         struct request_queue *q = bdev_get_queue(device->bdev);
3598         struct bio *bio = device->flush_bio;
3599
3600         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3601                 return;
3602
3603         bio_reset(bio);
3604         bio->bi_end_io = btrfs_end_empty_barrier;
3605         bio_set_dev(bio, device->bdev);
3606         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3607         init_completion(&device->flush_wait);
3608         bio->bi_private = &device->flush_wait;
3609
3610         btrfsic_submit_bio(bio);
3611         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3612 }
3613
3614 /*
3615  * If the flush bio has been submitted by write_dev_flush, wait for it.
3616  */
3617 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3618 {
3619         struct bio *bio = device->flush_bio;
3620
3621         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3622                 return BLK_STS_OK;
3623
3624         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3625         wait_for_completion_io(&device->flush_wait);
3626
3627         return bio->bi_status;
3628 }
3629
3630 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3631 {
3632         if (!btrfs_check_rw_degradable(fs_info, NULL))
3633                 return -EIO;
3634         return 0;
3635 }
3636
3637 /*
3638  * send an empty flush down to each device in parallel,
3639  * then wait for them
3640  */
3641 static int barrier_all_devices(struct btrfs_fs_info *info)
3642 {
3643         struct list_head *head;
3644         struct btrfs_device *dev;
3645         int errors_wait = 0;
3646         blk_status_t ret;
3647
3648         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3649         /* send down all the barriers */
3650         head = &info->fs_devices->devices;
3651         list_for_each_entry(dev, head, dev_list) {
3652                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3653                         continue;
3654                 if (!dev->bdev)
3655                         continue;
3656                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3657                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3658                         continue;
3659
3660                 write_dev_flush(dev);
3661                 dev->last_flush_error = BLK_STS_OK;
3662         }
3663
3664         /* wait for all the barriers */
3665         list_for_each_entry(dev, head, dev_list) {
3666                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3667                         continue;
3668                 if (!dev->bdev) {
3669                         errors_wait++;
3670                         continue;
3671                 }
3672                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3673                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3674                         continue;
3675
3676                 ret = wait_dev_flush(dev);
3677                 if (ret) {
3678                         dev->last_flush_error = ret;
3679                         btrfs_dev_stat_inc_and_print(dev,
3680                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3681                         errors_wait++;
3682                 }
3683         }
3684
3685         if (errors_wait) {
3686                 /*
3687                  * At some point we need the status of all disks
3688                  * to arrive at the volume status. So error checking
3689                  * is being pushed to a separate loop.
3690                  */
3691                 return check_barrier_error(info);
3692         }
3693         return 0;
3694 }
3695
3696 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3697 {
3698         int raid_type;
3699         int min_tolerated = INT_MAX;
3700
3701         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3702             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3703                 min_tolerated = min(min_tolerated,
3704                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3705                                     tolerated_failures);
3706
3707         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3708                 if (raid_type == BTRFS_RAID_SINGLE)
3709                         continue;
3710                 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3711                         continue;
3712                 min_tolerated = min(min_tolerated,
3713                                     btrfs_raid_array[raid_type].
3714                                     tolerated_failures);
3715         }
3716
3717         if (min_tolerated == INT_MAX) {
3718                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3719                 min_tolerated = 0;
3720         }
3721
3722         return min_tolerated;
3723 }
3724
3725 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3726 {
3727         struct list_head *head;
3728         struct btrfs_device *dev;
3729         struct btrfs_super_block *sb;
3730         struct btrfs_dev_item *dev_item;
3731         int ret;
3732         int do_barriers;
3733         int max_errors;
3734         int total_errors = 0;
3735         u64 flags;
3736
3737         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3738
3739         /*
3740          * max_mirrors == 0 indicates we're from commit_transaction,
3741          * not from fsync where the tree roots in fs_info have not
3742          * been consistent on disk.
3743          */
3744         if (max_mirrors == 0)
3745                 backup_super_roots(fs_info);
3746
3747         sb = fs_info->super_for_commit;
3748         dev_item = &sb->dev_item;
3749
3750         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3751         head = &fs_info->fs_devices->devices;
3752         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3753
3754         if (do_barriers) {
3755                 ret = barrier_all_devices(fs_info);
3756                 if (ret) {
3757                         mutex_unlock(
3758                                 &fs_info->fs_devices->device_list_mutex);
3759                         btrfs_handle_fs_error(fs_info, ret,
3760                                               "errors while submitting device barriers.");
3761                         return ret;
3762                 }
3763         }
3764
3765         list_for_each_entry(dev, head, dev_list) {
3766                 if (!dev->bdev) {
3767                         total_errors++;
3768                         continue;
3769                 }
3770                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3771                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3772                         continue;
3773
3774                 btrfs_set_stack_device_generation(dev_item, 0);
3775                 btrfs_set_stack_device_type(dev_item, dev->type);
3776                 btrfs_set_stack_device_id(dev_item, dev->devid);
3777                 btrfs_set_stack_device_total_bytes(dev_item,
3778                                                    dev->commit_total_bytes);
3779                 btrfs_set_stack_device_bytes_used(dev_item,
3780                                                   dev->commit_bytes_used);
3781                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3782                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3783                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3784                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3785                 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3786                        BTRFS_FSID_SIZE);
3787
3788                 flags = btrfs_super_flags(sb);
3789                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3790
3791                 ret = btrfs_validate_write_super(fs_info, sb);
3792                 if (ret < 0) {
3793                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3794                         btrfs_handle_fs_error(fs_info, -EUCLEAN,
3795                                 "unexpected superblock corruption detected");
3796                         return -EUCLEAN;
3797                 }
3798
3799                 ret = write_dev_supers(dev, sb, max_mirrors);
3800                 if (ret)
3801                         total_errors++;
3802         }
3803         if (total_errors > max_errors) {
3804                 btrfs_err(fs_info, "%d errors while writing supers",
3805                           total_errors);
3806                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3807
3808                 /* FUA is masked off if unsupported and can't be the reason */
3809                 btrfs_handle_fs_error(fs_info, -EIO,
3810                                       "%d errors while writing supers",
3811                                       total_errors);
3812                 return -EIO;
3813         }
3814
3815         total_errors = 0;
3816         list_for_each_entry(dev, head, dev_list) {
3817                 if (!dev->bdev)
3818                         continue;
3819                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3820                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3821                         continue;
3822
3823                 ret = wait_dev_supers(dev, max_mirrors);
3824                 if (ret)
3825                         total_errors++;
3826         }
3827         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3828         if (total_errors > max_errors) {
3829                 btrfs_handle_fs_error(fs_info, -EIO,
3830                                       "%d errors while writing supers",
3831                                       total_errors);
3832                 return -EIO;
3833         }
3834         return 0;
3835 }
3836
3837 /* Drop a fs root from the radix tree and free it. */
3838 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3839                                   struct btrfs_root *root)
3840 {
3841         spin_lock(&fs_info->fs_roots_radix_lock);
3842         radix_tree_delete(&fs_info->fs_roots_radix,
3843                           (unsigned long)root->root_key.objectid);
3844         spin_unlock(&fs_info->fs_roots_radix_lock);
3845
3846         if (btrfs_root_refs(&root->root_item) == 0)
3847                 synchronize_srcu(&fs_info->subvol_srcu);
3848
3849         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3850                 btrfs_free_log(NULL, root);
3851                 if (root->reloc_root) {
3852                         free_extent_buffer(root->reloc_root->node);
3853                         free_extent_buffer(root->reloc_root->commit_root);
3854                         btrfs_put_fs_root(root->reloc_root);
3855                         root->reloc_root = NULL;
3856                 }
3857         }
3858
3859         if (root->free_ino_pinned)
3860                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3861         if (root->free_ino_ctl)
3862                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3863         btrfs_free_fs_root(root);
3864 }
3865
3866 void btrfs_free_fs_root(struct btrfs_root *root)
3867 {
3868         iput(root->ino_cache_inode);
3869         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3870         if (root->anon_dev)
3871                 free_anon_bdev(root->anon_dev);
3872         if (root->subv_writers)
3873                 btrfs_free_subvolume_writers(root->subv_writers);
3874         free_extent_buffer(root->node);
3875         free_extent_buffer(root->commit_root);
3876         kfree(root->free_ino_ctl);
3877         kfree(root->free_ino_pinned);
3878         btrfs_put_fs_root(root);
3879 }
3880
3881 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3882 {
3883         u64 root_objectid = 0;
3884         struct btrfs_root *gang[8];
3885         int i = 0;
3886         int err = 0;
3887         unsigned int ret = 0;
3888         int index;
3889
3890         while (1) {
3891                 index = srcu_read_lock(&fs_info->subvol_srcu);
3892                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3893                                              (void **)gang, root_objectid,
3894                                              ARRAY_SIZE(gang));
3895                 if (!ret) {
3896                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3897                         break;
3898                 }
3899                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3900
3901                 for (i = 0; i < ret; i++) {
3902                         /* Avoid to grab roots in dead_roots */
3903                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3904                                 gang[i] = NULL;
3905                                 continue;
3906                         }
3907                         /* grab all the search result for later use */
3908                         gang[i] = btrfs_grab_fs_root(gang[i]);
3909                 }
3910                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3911
3912                 for (i = 0; i < ret; i++) {
3913                         if (!gang[i])
3914                                 continue;
3915                         root_objectid = gang[i]->root_key.objectid;
3916                         err = btrfs_orphan_cleanup(gang[i]);
3917                         if (err)
3918                                 break;
3919                         btrfs_put_fs_root(gang[i]);
3920                 }
3921                 root_objectid++;
3922         }
3923
3924         /* release the uncleaned roots due to error */
3925         for (; i < ret; i++) {
3926                 if (gang[i])
3927                         btrfs_put_fs_root(gang[i]);
3928         }
3929         return err;
3930 }
3931
3932 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3933 {
3934         struct btrfs_root *root = fs_info->tree_root;
3935         struct btrfs_trans_handle *trans;
3936
3937         mutex_lock(&fs_info->cleaner_mutex);
3938         btrfs_run_delayed_iputs(fs_info);
3939         mutex_unlock(&fs_info->cleaner_mutex);
3940         wake_up_process(fs_info->cleaner_kthread);
3941
3942         /* wait until ongoing cleanup work done */
3943         down_write(&fs_info->cleanup_work_sem);
3944         up_write(&fs_info->cleanup_work_sem);
3945
3946         trans = btrfs_join_transaction(root);
3947         if (IS_ERR(trans))
3948                 return PTR_ERR(trans);
3949         return btrfs_commit_transaction(trans);
3950 }
3951
3952 void close_ctree(struct btrfs_fs_info *fs_info)
3953 {
3954         int ret;
3955
3956         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3957         /*
3958          * We don't want the cleaner to start new transactions, add more delayed
3959          * iputs, etc. while we're closing. We can't use kthread_stop() yet
3960          * because that frees the task_struct, and the transaction kthread might
3961          * still try to wake up the cleaner.
3962          */
3963         kthread_park(fs_info->cleaner_kthread);
3964
3965         /* wait for the qgroup rescan worker to stop */
3966         btrfs_qgroup_wait_for_completion(fs_info, false);
3967
3968         /* wait for the uuid_scan task to finish */
3969         down(&fs_info->uuid_tree_rescan_sem);
3970         /* avoid complains from lockdep et al., set sem back to initial state */
3971         up(&fs_info->uuid_tree_rescan_sem);
3972
3973         /* pause restriper - we want to resume on mount */
3974         btrfs_pause_balance(fs_info);
3975
3976         btrfs_dev_replace_suspend_for_unmount(fs_info);
3977
3978         btrfs_scrub_cancel(fs_info);
3979
3980         /* wait for any defraggers to finish */
3981         wait_event(fs_info->transaction_wait,
3982                    (atomic_read(&fs_info->defrag_running) == 0));
3983
3984         /* clear out the rbtree of defraggable inodes */
3985         btrfs_cleanup_defrag_inodes(fs_info);
3986
3987         cancel_work_sync(&fs_info->async_reclaim_work);
3988
3989         if (!sb_rdonly(fs_info->sb)) {
3990                 /*
3991                  * The cleaner kthread is stopped, so do one final pass over
3992                  * unused block groups.
3993                  */
3994                 btrfs_delete_unused_bgs(fs_info);
3995
3996                 ret = btrfs_commit_super(fs_info);
3997                 if (ret)
3998                         btrfs_err(fs_info, "commit super ret %d", ret);
3999         }
4000
4001         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4002             test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4003                 btrfs_error_commit_super(fs_info);
4004
4005         kthread_stop(fs_info->transaction_kthread);
4006         kthread_stop(fs_info->cleaner_kthread);
4007
4008         ASSERT(list_empty(&fs_info->delayed_iputs));
4009         set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4010
4011         btrfs_free_qgroup_config(fs_info);
4012         ASSERT(list_empty(&fs_info->delalloc_roots));
4013
4014         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4015                 btrfs_info(fs_info, "at unmount delalloc count %lld",
4016                        percpu_counter_sum(&fs_info->delalloc_bytes));
4017         }
4018
4019         btrfs_sysfs_remove_mounted(fs_info);
4020         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4021
4022         btrfs_free_fs_roots(fs_info);
4023
4024         btrfs_put_block_group_cache(fs_info);
4025
4026         /*
4027          * we must make sure there is not any read request to
4028          * submit after we stopping all workers.
4029          */
4030         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4031         btrfs_stop_all_workers(fs_info);
4032
4033         btrfs_free_block_groups(fs_info);
4034
4035         clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4036         free_root_pointers(fs_info, 1);
4037
4038         iput(fs_info->btree_inode);
4039
4040 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4041         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4042                 btrfsic_unmount(fs_info->fs_devices);
4043 #endif
4044
4045         btrfs_close_devices(fs_info->fs_devices);
4046         btrfs_mapping_tree_free(&fs_info->mapping_tree);
4047
4048         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4049         percpu_counter_destroy(&fs_info->delalloc_bytes);
4050         percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
4051         cleanup_srcu_struct(&fs_info->subvol_srcu);
4052
4053         btrfs_free_stripe_hash_table(fs_info);
4054         btrfs_free_ref_cache(fs_info);
4055
4056         while (!list_empty(&fs_info->pinned_chunks)) {
4057                 struct extent_map *em;
4058
4059                 em = list_first_entry(&fs_info->pinned_chunks,
4060                                       struct extent_map, list);
4061                 list_del_init(&em->list);
4062                 free_extent_map(em);
4063         }
4064 }
4065
4066 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4067                           int atomic)
4068 {
4069         int ret;
4070         struct inode *btree_inode = buf->pages[0]->mapping->host;
4071
4072         ret = extent_buffer_uptodate(buf);
4073         if (!ret)
4074                 return ret;
4075
4076         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4077                                     parent_transid, atomic);
4078         if (ret == -EAGAIN)
4079                 return ret;
4080         return !ret;
4081 }
4082
4083 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4084 {
4085         struct btrfs_fs_info *fs_info;
4086         struct btrfs_root *root;
4087         u64 transid = btrfs_header_generation(buf);
4088         int was_dirty;
4089
4090 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4091         /*
4092          * This is a fast path so only do this check if we have sanity tests
4093          * enabled.  Normal people shouldn't be using unmapped buffers as dirty
4094          * outside of the sanity tests.
4095          */
4096         if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4097                 return;
4098 #endif
4099         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4100         fs_info = root->fs_info;
4101         btrfs_assert_tree_locked(buf);
4102         if (transid != fs_info->generation)
4103                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4104                         buf->start, transid, fs_info->generation);
4105         was_dirty = set_extent_buffer_dirty(buf);
4106         if (!was_dirty)
4107                 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4108                                          buf->len,
4109                                          fs_info->dirty_metadata_batch);
4110 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4111         /*
4112          * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4113          * but item data not updated.
4114          * So here we should only check item pointers, not item data.
4115          */
4116         if (btrfs_header_level(buf) == 0 &&
4117             btrfs_check_leaf_relaxed(fs_info, buf)) {
4118                 btrfs_print_leaf(buf);
4119                 ASSERT(0);
4120         }
4121 #endif
4122 }
4123
4124 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4125                                         int flush_delayed)
4126 {
4127         /*
4128          * looks as though older kernels can get into trouble with
4129          * this code, they end up stuck in balance_dirty_pages forever
4130          */
4131         int ret;
4132
4133         if (current->flags & PF_MEMALLOC)
4134                 return;
4135
4136         if (flush_delayed)
4137                 btrfs_balance_delayed_items(fs_info);
4138
4139         ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4140                                      BTRFS_DIRTY_METADATA_THRESH,
4141                                      fs_info->dirty_metadata_batch);
4142         if (ret > 0) {
4143                 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4144         }
4145 }
4146
4147 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4148 {
4149         __btrfs_btree_balance_dirty(fs_info, 1);
4150 }
4151
4152 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4153 {
4154         __btrfs_btree_balance_dirty(fs_info, 0);
4155 }
4156
4157 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4158                       struct btrfs_key *first_key)
4159 {
4160         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4161         struct btrfs_fs_info *fs_info = root->fs_info;
4162
4163         return btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
4164                                               level, first_key);
4165 }
4166
4167 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4168 {
4169         /* cleanup FS via transaction */
4170         btrfs_cleanup_transaction(fs_info);
4171
4172         mutex_lock(&fs_info->cleaner_mutex);
4173         btrfs_run_delayed_iputs(fs_info);
4174         mutex_unlock(&fs_info->cleaner_mutex);
4175
4176         down_write(&fs_info->cleanup_work_sem);
4177         up_write(&fs_info->cleanup_work_sem);
4178 }
4179
4180 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4181 {
4182         struct btrfs_ordered_extent *ordered;
4183
4184         spin_lock(&root->ordered_extent_lock);
4185         /*
4186          * This will just short circuit the ordered completion stuff which will
4187          * make sure the ordered extent gets properly cleaned up.
4188          */
4189         list_for_each_entry(ordered, &root->ordered_extents,
4190                             root_extent_list)
4191                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4192         spin_unlock(&root->ordered_extent_lock);
4193 }
4194
4195 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4196 {
4197         struct btrfs_root *root;
4198         struct list_head splice;
4199
4200         INIT_LIST_HEAD(&splice);
4201
4202         spin_lock(&fs_info->ordered_root_lock);
4203         list_splice_init(&fs_info->ordered_roots, &splice);
4204         while (!list_empty(&splice)) {
4205                 root = list_first_entry(&splice, struct btrfs_root,
4206                                         ordered_root);
4207                 list_move_tail(&root->ordered_root,
4208                                &fs_info->ordered_roots);
4209
4210                 spin_unlock(&fs_info->ordered_root_lock);
4211                 btrfs_destroy_ordered_extents(root);
4212
4213                 cond_resched();
4214                 spin_lock(&fs_info->ordered_root_lock);
4215         }
4216         spin_unlock(&fs_info->ordered_root_lock);
4217
4218         /*
4219          * We need this here because if we've been flipped read-only we won't
4220          * get sync() from the umount, so we need to make sure any ordered
4221          * extents that haven't had their dirty pages IO start writeout yet
4222          * actually get run and error out properly.
4223          */
4224         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4225 }
4226
4227 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4228                                       struct btrfs_fs_info *fs_info)
4229 {
4230         struct rb_node *node;
4231         struct btrfs_delayed_ref_root *delayed_refs;
4232         struct btrfs_delayed_ref_node *ref;
4233         int ret = 0;
4234
4235         delayed_refs = &trans->delayed_refs;
4236
4237         spin_lock(&delayed_refs->lock);
4238         if (atomic_read(&delayed_refs->num_entries) == 0) {
4239                 spin_unlock(&delayed_refs->lock);
4240                 btrfs_info(fs_info, "delayed_refs has NO entry");
4241                 return ret;
4242         }
4243
4244         while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4245                 struct btrfs_delayed_ref_head *head;
4246                 struct rb_node *n;
4247                 bool pin_bytes = false;
4248
4249                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4250                                 href_node);
4251                 if (btrfs_delayed_ref_lock(delayed_refs, head))
4252                         continue;
4253
4254                 spin_lock(&head->lock);
4255                 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4256                         ref = rb_entry(n, struct btrfs_delayed_ref_node,
4257                                        ref_node);
4258                         ref->in_tree = 0;
4259                         rb_erase_cached(&ref->ref_node, &head->ref_tree);
4260                         RB_CLEAR_NODE(&ref->ref_node);
4261                         if (!list_empty(&ref->add_list))
4262                                 list_del(&ref->add_list);
4263                         atomic_dec(&delayed_refs->num_entries);
4264                         btrfs_put_delayed_ref(ref);
4265                 }
4266                 if (head->must_insert_reserved)
4267                         pin_bytes = true;
4268                 btrfs_free_delayed_extent_op(head->extent_op);
4269                 btrfs_delete_ref_head(delayed_refs, head);
4270                 spin_unlock(&head->lock);
4271                 spin_unlock(&delayed_refs->lock);
4272                 mutex_unlock(&head->mutex);
4273
4274                 if (pin_bytes)
4275                         btrfs_pin_extent(fs_info, head->bytenr,
4276                                          head->num_bytes, 1);
4277                 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4278                 btrfs_put_delayed_ref_head(head);
4279                 cond_resched();
4280                 spin_lock(&delayed_refs->lock);
4281         }
4282
4283         spin_unlock(&delayed_refs->lock);
4284
4285         return ret;
4286 }
4287
4288 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4289 {
4290         struct btrfs_inode *btrfs_inode;
4291         struct list_head splice;
4292
4293         INIT_LIST_HEAD(&splice);
4294
4295         spin_lock(&root->delalloc_lock);
4296         list_splice_init(&root->delalloc_inodes, &splice);
4297
4298         while (!list_empty(&splice)) {
4299                 struct inode *inode = NULL;
4300                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4301                                                delalloc_inodes);
4302                 __btrfs_del_delalloc_inode(root, btrfs_inode);
4303                 spin_unlock(&root->delalloc_lock);
4304
4305                 /*
4306                  * Make sure we get a live inode and that it'll not disappear
4307                  * meanwhile.
4308                  */
4309                 inode = igrab(&btrfs_inode->vfs_inode);
4310                 if (inode) {
4311                         invalidate_inode_pages2(inode->i_mapping);
4312                         iput(inode);
4313                 }
4314                 spin_lock(&root->delalloc_lock);
4315         }
4316         spin_unlock(&root->delalloc_lock);
4317 }
4318
4319 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4320 {
4321         struct btrfs_root *root;
4322         struct list_head splice;
4323
4324         INIT_LIST_HEAD(&splice);
4325
4326         spin_lock(&fs_info->delalloc_root_lock);
4327         list_splice_init(&fs_info->delalloc_roots, &splice);
4328         while (!list_empty(&splice)) {
4329                 root = list_first_entry(&splice, struct btrfs_root,
4330                                          delalloc_root);
4331                 root = btrfs_grab_fs_root(root);
4332                 BUG_ON(!root);
4333                 spin_unlock(&fs_info->delalloc_root_lock);
4334
4335                 btrfs_destroy_delalloc_inodes(root);
4336                 btrfs_put_fs_root(root);
4337
4338                 spin_lock(&fs_info->delalloc_root_lock);
4339         }
4340         spin_unlock(&fs_info->delalloc_root_lock);
4341 }
4342
4343 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4344                                         struct extent_io_tree *dirty_pages,
4345                                         int mark)
4346 {
4347         int ret;
4348         struct extent_buffer *eb;
4349         u64 start = 0;
4350         u64 end;
4351
4352         while (1) {
4353                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4354                                             mark, NULL);
4355                 if (ret)
4356                         break;
4357
4358                 clear_extent_bits(dirty_pages, start, end, mark);
4359                 while (start <= end) {
4360                         eb = find_extent_buffer(fs_info, start);
4361                         start += fs_info->nodesize;
4362                         if (!eb)
4363                                 continue;
4364                         wait_on_extent_buffer_writeback(eb);
4365
4366                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4367                                                &eb->bflags))
4368                                 clear_extent_buffer_dirty(eb);
4369                         free_extent_buffer_stale(eb);
4370                 }
4371         }
4372
4373         return ret;
4374 }
4375
4376 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4377                                        struct extent_io_tree *pinned_extents)
4378 {
4379         struct extent_io_tree *unpin;
4380         u64 start;
4381         u64 end;
4382         int ret;
4383         bool loop = true;
4384
4385         unpin = pinned_extents;
4386 again:
4387         while (1) {
4388                 struct extent_state *cached_state = NULL;
4389
4390                 /*
4391                  * The btrfs_finish_extent_commit() may get the same range as
4392                  * ours between find_first_extent_bit and clear_extent_dirty.
4393                  * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4394                  * the same extent range.
4395                  */
4396                 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4397                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4398                                             EXTENT_DIRTY, &cached_state);
4399                 if (ret) {
4400                         mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4401                         break;
4402                 }
4403
4404                 clear_extent_dirty(unpin, start, end, &cached_state);
4405                 free_extent_state(cached_state);
4406                 btrfs_error_unpin_extent_range(fs_info, start, end);
4407                 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4408                 cond_resched();
4409         }
4410
4411         if (loop) {
4412                 if (unpin == &fs_info->freed_extents[0])
4413                         unpin = &fs_info->freed_extents[1];
4414                 else
4415                         unpin = &fs_info->freed_extents[0];
4416                 loop = false;
4417                 goto again;
4418         }
4419
4420         return 0;
4421 }
4422
4423 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4424 {
4425         struct inode *inode;
4426
4427         inode = cache->io_ctl.inode;
4428         if (inode) {
4429                 invalidate_inode_pages2(inode->i_mapping);
4430                 BTRFS_I(inode)->generation = 0;
4431                 cache->io_ctl.inode = NULL;
4432                 iput(inode);
4433         }
4434         btrfs_put_block_group(cache);
4435 }
4436
4437 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4438                              struct btrfs_fs_info *fs_info)
4439 {
4440         struct btrfs_block_group_cache *cache;
4441
4442         spin_lock(&cur_trans->dirty_bgs_lock);
4443         while (!list_empty(&cur_trans->dirty_bgs)) {
4444                 cache = list_first_entry(&cur_trans->dirty_bgs,
4445                                          struct btrfs_block_group_cache,
4446                                          dirty_list);
4447
4448                 if (!list_empty(&cache->io_list)) {
4449                         spin_unlock(&cur_trans->dirty_bgs_lock);
4450                         list_del_init(&cache->io_list);
4451                         btrfs_cleanup_bg_io(cache);
4452                         spin_lock(&cur_trans->dirty_bgs_lock);
4453                 }
4454
4455                 list_del_init(&cache->dirty_list);
4456                 spin_lock(&cache->lock);
4457                 cache->disk_cache_state = BTRFS_DC_ERROR;
4458                 spin_unlock(&cache->lock);
4459
4460                 spin_unlock(&cur_trans->dirty_bgs_lock);
4461                 btrfs_put_block_group(cache);
4462                 btrfs_delayed_refs_rsv_release(fs_info, 1);
4463                 spin_lock(&cur_trans->dirty_bgs_lock);
4464         }
4465         spin_unlock(&cur_trans->dirty_bgs_lock);
4466
4467         /*
4468          * Refer to the definition of io_bgs member for details why it's safe
4469          * to use it without any locking
4470          */
4471         while (!list_empty(&cur_trans->io_bgs)) {
4472                 cache = list_first_entry(&cur_trans->io_bgs,
4473                                          struct btrfs_block_group_cache,
4474                                          io_list);
4475
4476                 list_del_init(&cache->io_list);
4477                 spin_lock(&cache->lock);
4478                 cache->disk_cache_state = BTRFS_DC_ERROR;
4479                 spin_unlock(&cache->lock);
4480                 btrfs_cleanup_bg_io(cache);
4481         }
4482 }
4483
4484 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4485                                    struct btrfs_fs_info *fs_info)
4486 {
4487         btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4488         ASSERT(list_empty(&cur_trans->dirty_bgs));
4489         ASSERT(list_empty(&cur_trans->io_bgs));
4490
4491         btrfs_destroy_delayed_refs(cur_trans, fs_info);
4492
4493         cur_trans->state = TRANS_STATE_COMMIT_START;
4494         wake_up(&fs_info->transaction_blocked_wait);
4495
4496         cur_trans->state = TRANS_STATE_UNBLOCKED;
4497         wake_up(&fs_info->transaction_wait);
4498
4499         btrfs_destroy_delayed_inodes(fs_info);
4500         btrfs_assert_delayed_root_empty(fs_info);
4501
4502         btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4503                                      EXTENT_DIRTY);
4504         btrfs_destroy_pinned_extent(fs_info,
4505                                     fs_info->pinned_extents);
4506
4507         cur_trans->state =TRANS_STATE_COMPLETED;
4508         wake_up(&cur_trans->commit_wait);
4509 }
4510
4511 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4512 {
4513         struct btrfs_transaction *t;
4514
4515         mutex_lock(&fs_info->transaction_kthread_mutex);
4516
4517         spin_lock(&fs_info->trans_lock);
4518         while (!list_empty(&fs_info->trans_list)) {
4519                 t = list_first_entry(&fs_info->trans_list,
4520                                      struct btrfs_transaction, list);
4521                 if (t->state >= TRANS_STATE_COMMIT_START) {
4522                         refcount_inc(&t->use_count);
4523                         spin_unlock(&fs_info->trans_lock);
4524                         btrfs_wait_for_commit(fs_info, t->transid);
4525                         btrfs_put_transaction(t);
4526                         spin_lock(&fs_info->trans_lock);
4527                         continue;
4528                 }
4529                 if (t == fs_info->running_transaction) {
4530                         t->state = TRANS_STATE_COMMIT_DOING;
4531                         spin_unlock(&fs_info->trans_lock);
4532                         /*
4533                          * We wait for 0 num_writers since we don't hold a trans
4534                          * handle open currently for this transaction.
4535                          */
4536                         wait_event(t->writer_wait,
4537                                    atomic_read(&t->num_writers) == 0);
4538                 } else {
4539                         spin_unlock(&fs_info->trans_lock);
4540                 }
4541                 btrfs_cleanup_one_transaction(t, fs_info);
4542
4543                 spin_lock(&fs_info->trans_lock);
4544                 if (t == fs_info->running_transaction)
4545                         fs_info->running_transaction = NULL;
4546                 list_del_init(&t->list);
4547                 spin_unlock(&fs_info->trans_lock);
4548
4549                 btrfs_put_transaction(t);
4550                 trace_btrfs_transaction_commit(fs_info->tree_root);
4551                 spin_lock(&fs_info->trans_lock);
4552         }
4553         spin_unlock(&fs_info->trans_lock);
4554         btrfs_destroy_all_ordered_extents(fs_info);
4555         btrfs_destroy_delayed_inodes(fs_info);
4556         btrfs_assert_delayed_root_empty(fs_info);
4557         btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
4558         btrfs_destroy_all_delalloc_inodes(fs_info);
4559         mutex_unlock(&fs_info->transaction_kthread_mutex);
4560
4561         return 0;
4562 }
4563
4564 static const struct extent_io_ops btree_extent_io_ops = {
4565         /* mandatory callbacks */
4566         .submit_bio_hook = btree_submit_bio_hook,
4567         .readpage_end_io_hook = btree_readpage_end_io_hook,
4568 };